drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    This program is free software; you can redistribute it and/or modify
  23    it under the terms of the GNU General Public License as published by
  24    the Free Software Foundation; either version 2, or (at your option)
  25    any later version.
  26
  27    You should have received a copy of the GNU General Public License
  28    (for example /usr/src/linux/COPYING); if not, write to the Free
  29    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  30 */
  31
  32 #include <linux/module.h>
  33 #include <linux/config.h>
  34 #include <linux/linkage.h>
  35 #include <linux/raid/md.h>
  36 #include <linux/sysctl.h>
  37 #include <linux/devfs_fs_kernel.h>
  38 #include <linux/buffer_head.h> /* for invalidate_bdev */
  39 #include <linux/suspend.h>
  40
  41 #include <linux/init.h>
  42
  43 #ifdef CONFIG_KMOD
  44 #include <linux/kmod.h>
  45 #endif
  46
  47 #include <asm/unaligned.h>
  48
  49 #define MAJOR_NR MD_MAJOR
  50 #define MD_DRIVER
  51
  52 /* 63 partitions with the alternate major number (mdp) */
  53 #define MdpMinorShift 6
  54
  55 #define DEBUG 0
  56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  57
  58
  59 #ifndef MODULE
  60 static void autostart_arrays (int part);
  61 #endif
  62
  63 static mdk_personality_t *pers[MAX_PERSONALITY];
  64 static spinlock_t pers_lock = SPIN_LOCK_UNLOCKED;
  65
  66 /*
  67  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  68  * is 1000 KB/sec, so the extra system load does not show up that much.
  69  * Increase it if you want to have more _guaranteed_ speed. Note that
  70  * the RAID driver will use the maximum available bandwith if the IO
  71  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  72  * speed limit - in case reconstruction slows down your system despite
  73  * idle IO detection.
  74  *
  75  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  76  */
  77
  78 static int sysctl_speed_limit_min = 1000;
  79 static int sysctl_speed_limit_max = 200000;
  80
  81 static struct ctl_table_header *raid_table_header;
  82
  83 static ctl_table raid_table[] = {
  84         {
  85                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  86                 .procname       = "speed_limit_min",
  87                 .data           = &sysctl_speed_limit_min,
  88                 .maxlen         = sizeof(int),
  89                 .mode           = 0644,
  90                 .proc_handler   = &proc_dointvec,
  91         },
  92         {
  93                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
  94                 .procname       = "speed_limit_max",
  95                 .data           = &sysctl_speed_limit_max,
  96                 .maxlen         = sizeof(int),
  97                 .mode           = 0644,
  98                 .proc_handler   = &proc_dointvec,
  99         },
 100         { .ctl_name = 0 }
 101 };
 102
 103 static ctl_table raid_dir_table[] = {
 104         {
 105                 .ctl_name       = DEV_RAID,
 106                 .procname       = "raid",
 107                 .maxlen         = 0,
 108                 .mode           = 0555,
 109                 .child          = raid_table,
 110         },
 111         { .ctl_name = 0 }
 112 };
 113
 114 static ctl_table raid_root_table[] = {
 115         {
 116                 .ctl_name       = CTL_DEV,
 117                 .procname       = "dev",
 118                 .maxlen         = 0,
 119                 .mode           = 0555,
 120                 .child          = raid_dir_table,
 121         },
 122         { .ctl_name = 0 }
 123 };
 124
 125 static struct block_device_operations md_fops;
 126
 127 /*
 128  * Enables to iterate over all existing md arrays
 129  * all_mddevs_lock protects this list.
 130  */
 131 static LIST_HEAD(all_mddevs);
 132 static spinlock_t all_mddevs_lock = SPIN_LOCK_UNLOCKED;
 133
 134
 135 /*
 136  * iterates through all used mddevs in the system.
 137  * We take care to grab the all_mddevs_lock whenever navigating
 138  * the list, and to always hold a refcount when unlocked.
 139  * Any code which breaks out of this loop while own
 140  * a reference to the current mddev and must mddev_put it.
 141  */
 142 #define ITERATE_MDDEV(mddev,tmp)                                        \
 143                                                                         \
 144         for (({ spin_lock(&all_mddevs_lock);                            \
 145                 tmp = all_mddevs.next;                                  \
 146                 mddev = NULL;});                                        \
 147              ({ if (tmp != &all_mddevs)                                 \
 148                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 149                 spin_unlock(&all_mddevs_lock);                          \
 150                 if (mddev) mddev_put(mddev);                            \
 151                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 152                 tmp != &all_mddevs;});                                  \
 153              ({ spin_lock(&all_mddevs_lock);                            \
 154                 tmp = tmp->next;})                                      \
 155                 )
 156
 157 static int md_fail_request (request_queue_t *q, struct bio *bio)
 158 {
 159         bio_io_error(bio, bio->bi_size);
 160         return 0;
 161 }
 162
 163 static inline mddev_t *mddev_get(mddev_t *mddev)
 164 {
 165         atomic_inc(&mddev->active);
 166         return mddev;
 167 }
 168
 169 static void mddev_put(mddev_t *mddev)
 170 {
 171         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 172                 return;
 173         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 174                 list_del(&mddev->all_mddevs);
 175                 blk_put_queue(mddev->queue);
 176                 kfree(mddev);
 177         }
 178         spin_unlock(&all_mddevs_lock);
 179 }
 180
 181 static mddev_t * mddev_find(dev_t unit)
 182 {
 183         mddev_t *mddev, *new = NULL;
 184
 185  retry:
 186         spin_lock(&all_mddevs_lock);
 187         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 188                 if (mddev->unit == unit) {
 189                         mddev_get(mddev);
 190                         spin_unlock(&all_mddevs_lock);
 191                         if (new)
 192                                 kfree(new);
 193                         return mddev;
 194                 }
 195
 196         if (new) {
 197                 list_add(&new->all_mddevs, &all_mddevs);
 198                 spin_unlock(&all_mddevs_lock);
 199                 return new;
 200         }
 201         spin_unlock(&all_mddevs_lock);
 202
 203         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 204         if (!new)
 205                 return NULL;
 206
 207         memset(new, 0, sizeof(*new));
 208
 209         new->unit = unit;
 210         if (MAJOR(unit) == MD_MAJOR)
 211                 new->md_minor = MINOR(unit);
 212         else
 213                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 214
 215         init_MUTEX(&new->reconfig_sem);
 216         INIT_LIST_HEAD(&new->disks);
 217         INIT_LIST_HEAD(&new->all_mddevs);
 218         init_timer(&new->safemode_timer);
 219         atomic_set(&new->active, 1);
 220
 221         new->queue = blk_alloc_queue(GFP_KERNEL);
 222         if (!new->queue) {
 223                 kfree(new);
 224                 return NULL;
 225         }
 226
 227         blk_queue_make_request(new->queue, md_fail_request);
 228
 229         goto retry;
 230 }
 231
 232 static inline int mddev_lock(mddev_t * mddev)
 233 {
 234         return down_interruptible(&mddev->reconfig_sem);
 235 }
 236
 237 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 238 {
 239         down(&mddev->reconfig_sem);
 240 }
 241
 242 static inline int mddev_trylock(mddev_t * mddev)
 243 {
 244         return down_trylock(&mddev->reconfig_sem);
 245 }
 246
 247 static inline void mddev_unlock(mddev_t * mddev)
 248 {
 249         up(&mddev->reconfig_sem);
 250
 251         if (mddev->thread)
 252                 md_wakeup_thread(mddev->thread);
 253 }
 254
 255 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 256 {
 257         mdk_rdev_t * rdev;
 258         struct list_head *tmp;
 259
 260         ITERATE_RDEV(mddev,rdev,tmp) {
 261                 if (rdev->desc_nr == nr)
 262                         return rdev;
 263         }
 264         return NULL;
 265 }
 266
 267 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 268 {
 269         struct list_head *tmp;
 270         mdk_rdev_t *rdev;
 271
 272         ITERATE_RDEV(mddev,rdev,tmp) {
 273                 if (rdev->bdev->bd_dev == dev)
 274                         return rdev;
 275         }
 276         return NULL;
 277 }
 278
 279 inline static sector_t calc_dev_sboffset(struct block_device *bdev)
 280 {
 281         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 282         return MD_NEW_SIZE_BLOCKS(size);
 283 }
 284
 285 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 286 {
 287         sector_t size;
 288
 289         size = rdev->sb_offset;
 290
 291         if (chunk_size)
 292                 size &= ~((sector_t)chunk_size/1024 - 1);
 293         return size;
 294 }
 295
 296 static int alloc_disk_sb(mdk_rdev_t * rdev)
 297 {
 298         if (rdev->sb_page)
 299                 MD_BUG();
 300
 301         rdev->sb_page = alloc_page(GFP_KERNEL);
 302         if (!rdev->sb_page) {
 303                 printk(KERN_ALERT "md: out of memory.\n");
 304                 return -EINVAL;
 305         }
 306
 307         return 0;
 308 }
 309
 310 static void free_disk_sb(mdk_rdev_t * rdev)
 311 {
 312         if (rdev->sb_page) {
 313                 page_cache_release(rdev->sb_page);
 314                 rdev->sb_loaded = 0;
 315                 rdev->sb_page = NULL;
 316                 rdev->sb_offset = 0;
 317                 rdev->size = 0;
 318         }
 319 }
 320
 321
 322 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 323 {
 324         if (bio->bi_size)
 325                 return 1;
 326
 327         complete((struct completion*)bio->bi_private);
 328         return 0;
 329 }
 330
 331 static int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 332                    struct page *page, int rw)
 333 {
 334         struct bio bio;
 335         struct bio_vec vec;
 336         struct completion event;
 337
 338         rw |= (1 << BIO_RW_SYNC);
 339
 340         bio_init(&bio);
 341         bio.bi_io_vec = &vec;
 342         vec.bv_page = page;
 343         vec.bv_len = size;
 344         vec.bv_offset = 0;
 345         bio.bi_vcnt = 1;
 346         bio.bi_idx = 0;
 347         bio.bi_size = size;
 348         bio.bi_bdev = bdev;
 349         bio.bi_sector = sector;
 350         init_completion(&event);
 351         bio.bi_private = &event;
 352         bio.bi_end_io = bi_complete;
 353         submit_bio(rw, &bio);
 354         wait_for_completion(&event);
 355
 356         return test_bit(BIO_UPTODATE, &bio.bi_flags);
 357 }
 358
 359 static int read_disk_sb(mdk_rdev_t * rdev)
 360 {
 361         char b[BDEVNAME_SIZE];
 362         if (!rdev->sb_page) {
 363                 MD_BUG();
 364                 return -EINVAL;
 365         }
 366         if (rdev->sb_loaded)
 367                 return 0;
 368
 369
 370         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
 371                 goto fail;
 372         rdev->sb_loaded = 1;
 373         return 0;
 374
 375 fail:
 376         printk(KERN_ERR "md: disabled device %s, could not read superblock.\n",
 377                 bdevname(rdev->bdev,b));
 378         return -EINVAL;
 379 }
 380
 381 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 382 {
 383         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 384                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 385                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 386                 (sb1->set_uuid3 == sb2->set_uuid3))
 387
 388                 return 1;
 389
 390         return 0;
 391 }
 392
 393
 394 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 395 {
 396         int ret;
 397         mdp_super_t *tmp1, *tmp2;
 398
 399         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 400         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 401
 402         if (!tmp1 || !tmp2) {
 403                 ret = 0;
 404                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 405                 goto abort;
 406         }
 407
 408         *tmp1 = *sb1;
 409         *tmp2 = *sb2;
 410
 411         /*
 412          * nr_disks is not constant
 413          */
 414         tmp1->nr_disks = 0;
 415         tmp2->nr_disks = 0;
 416
 417         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 418                 ret = 0;
 419         else
 420                 ret = 1;
 421
 422 abort:
 423         if (tmp1)
 424                 kfree(tmp1);
 425         if (tmp2)
 426                 kfree(tmp2);
 427
 428         return ret;
 429 }
 430
 431 static unsigned int calc_sb_csum(mdp_super_t * sb)
 432 {
 433         unsigned int disk_csum, csum;
 434
 435         disk_csum = sb->sb_csum;
 436         sb->sb_csum = 0;
 437         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 438         sb->sb_csum = disk_csum;
 439         return csum;
 440 }
 441
 442 /*
 443  * Handle superblock details.
 444  * We want to be able to handle multiple superblock formats
 445  * so we have a common interface to them all, and an array of
 446  * different handlers.
 447  * We rely on user-space to write the initial superblock, and support
 448  * reading and updating of superblocks.
 449  * Interface methods are:
 450  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 451  *      loads and validates a superblock on dev.
 452  *      if refdev != NULL, compare superblocks on both devices
 453  *    Return:
 454  *      0 - dev has a superblock that is compatible with refdev
 455  *      1 - dev has a superblock that is compatible and newer than refdev
 456  *          so dev should be used as the refdev in future
 457  *     -EINVAL superblock incompatible or invalid
 458  *     -othererror e.g. -EIO
 459  *
 460  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 461  *      Verify that dev is acceptable into mddev.
 462  *       The first time, mddev->raid_disks will be 0, and data from
 463  *       dev should be merged in.  Subsequent calls check that dev
 464  *       is new enough.  Return 0 or -EINVAL
 465  *
 466  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 467  *     Update the superblock for rdev with data in mddev
 468  *     This does not write to disc.
 469  *
 470  */
 471
 472 struct super_type  {
 473         char            *name;
 474         struct module   *owner;
 475         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 476         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 477         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 478 };
 479
 480 /*
 481  * load_super for 0.90.0
 482  */
 483 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 484 {
 485         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 486         mdp_super_t *sb;
 487         int ret;
 488         sector_t sb_offset;
 489
 490         /*
 491          * Calculate the position of the superblock,
 492          * it's at the end of the disk.
 493          *
 494          * It also happens to be a multiple of 4Kb.
 495          */
 496         sb_offset = calc_dev_sboffset(rdev->bdev);
 497         rdev->sb_offset = sb_offset;
 498
 499         ret = read_disk_sb(rdev);
 500         if (ret) return ret;
 501
 502         ret = -EINVAL;
 503
 504         bdevname(rdev->bdev, b);
 505         sb = (mdp_super_t*)page_address(rdev->sb_page);
 506
 507         if (sb->md_magic != MD_SB_MAGIC) {
 508                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 509                        b);
 510                 goto abort;
 511         }
 512
 513         if (sb->major_version != 0 ||
 514             sb->minor_version != 90) {
 515                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 516                         sb->major_version, sb->minor_version,
 517                         b);
 518                 goto abort;
 519         }
 520
 521         if (sb->raid_disks <= 0)
 522                 goto abort;
 523
 524         if (calc_sb_csum(sb) != sb->sb_csum) {
 525                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 526                         b);
 527                 goto abort;
 528         }
 529
 530         rdev->preferred_minor = sb->md_minor;
 531         rdev->data_offset = 0;
 532
 533         if (sb->level == MULTIPATH)
 534                 rdev->desc_nr = -1;
 535         else
 536                 rdev->desc_nr = sb->this_disk.number;
 537
 538         if (refdev == 0)
 539                 ret = 1;
 540         else {
 541                 __u64 ev1, ev2;
 542                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 543                 if (!uuid_equal(refsb, sb)) {
 544                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 545                                 b, bdevname(refdev->bdev,b2));
 546                         goto abort;
 547                 }
 548                 if (!sb_equal(refsb, sb)) {
 549                         printk(KERN_WARNING "md: %s has same UUID"
 550                                " but different superblock to %s\n",
 551                                b, bdevname(refdev->bdev, b2));
 552                         goto abort;
 553                 }
 554                 ev1 = md_event(sb);
 555                 ev2 = md_event(refsb);
 556                 if (ev1 > ev2)
 557                         ret = 1;
 558                 else
 559                         ret = 0;
 560         }
 561         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 562
 563  abort:
 564         return ret;
 565 }
 566
 567 /*
 568  * validate_super for 0.90.0
 569  */
 570 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 571 {
 572         mdp_disk_t *desc;
 573         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 574
 575         if (mddev->raid_disks == 0) {
 576                 mddev->major_version = 0;
 577                 mddev->minor_version = sb->minor_version;
 578                 mddev->patch_version = sb->patch_version;
 579                 mddev->persistent = ! sb->not_persistent;
 580                 mddev->chunk_size = sb->chunk_size;
 581                 mddev->ctime = sb->ctime;
 582                 mddev->utime = sb->utime;
 583                 mddev->level = sb->level;
 584                 mddev->layout = sb->layout;
 585                 mddev->raid_disks = sb->raid_disks;
 586                 mddev->size = sb->size;
 587                 mddev->events = md_event(sb);
 588
 589                 if (sb->state & (1<<MD_SB_CLEAN))
 590                         mddev->recovery_cp = MaxSector;
 591                 else {
 592                         if (sb->events_hi == sb->cp_events_hi &&
 593                                 sb->events_lo == sb->cp_events_lo) {
 594                                 mddev->recovery_cp = sb->recovery_cp;
 595                         } else
 596                                 mddev->recovery_cp = 0;
 597                 }
 598
 599                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 600                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 601                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 602                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 603
 604                 mddev->max_disks = MD_SB_DISKS;
 605         } else {
 606                 __u64 ev1;
 607                 ev1 = md_event(sb);
 608                 ++ev1;
 609                 if (ev1 < mddev->events)
 610                         return -EINVAL;
 611         }
 612         if (mddev->level != LEVEL_MULTIPATH) {
 613                 rdev->raid_disk = -1;
 614                 rdev->in_sync = rdev->faulty = 0;
 615                 desc = sb->disks + rdev->desc_nr;
 616
 617                 if (desc->state & (1<<MD_DISK_FAULTY))
 618                         rdev->faulty = 1;
 619                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 620                          desc->raid_disk < mddev->raid_disks) {
 621                         rdev->in_sync = 1;
 622                         rdev->raid_disk = desc->raid_disk;
 623                 }
 624         }
 625         return 0;
 626 }
 627
 628 /*
 629  * sync_super for 0.90.0
 630  */
 631 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 632 {
 633         mdp_super_t *sb;
 634         struct list_head *tmp;
 635         mdk_rdev_t *rdev2;
 636         int next_spare = mddev->raid_disks;
 637
 638         /* make rdev->sb match mddev data..
 639          *
 640          * 1/ zero out disks
 641          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 642          * 3/ any empty disks < next_spare become removed
 643          *
 644          * disks[0] gets initialised to REMOVED because
 645          * we cannot be sure from other fields if it has
 646          * been initialised or not.
 647          */
 648         int i;
 649         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 650
 651         sb = (mdp_super_t*)page_address(rdev->sb_page);
 652
 653         memset(sb, 0, sizeof(*sb));
 654
 655         sb->md_magic = MD_SB_MAGIC;
 656         sb->major_version = mddev->major_version;
 657         sb->minor_version = mddev->minor_version;
 658         sb->patch_version = mddev->patch_version;
 659         sb->gvalid_words  = 0; /* ignored */
 660         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 661         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 662         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 663         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 664
 665         sb->ctime = mddev->ctime;
 666         sb->level = mddev->level;
 667         sb->size  = mddev->size;
 668         sb->raid_disks = mddev->raid_disks;
 669         sb->md_minor = mddev->md_minor;
 670         sb->not_persistent = !mddev->persistent;
 671         sb->utime = mddev->utime;
 672         sb->state = 0;
 673         sb->events_hi = (mddev->events>>32);
 674         sb->events_lo = (u32)mddev->events;
 675
 676         if (mddev->in_sync)
 677         {
 678                 sb->recovery_cp = mddev->recovery_cp;
 679                 sb->cp_events_hi = (mddev->events>>32);
 680                 sb->cp_events_lo = (u32)mddev->events;
 681                 if (mddev->recovery_cp == MaxSector)
 682                         sb->state = (1<< MD_SB_CLEAN);
 683         } else
 684                 sb->recovery_cp = 0;
 685
 686         sb->layout = mddev->layout;
 687         sb->chunk_size = mddev->chunk_size;
 688
 689         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 690         ITERATE_RDEV(mddev,rdev2,tmp) {
 691                 mdp_disk_t *d;
 692                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 693                         rdev2->desc_nr = rdev2->raid_disk;
 694                 else
 695                         rdev2->desc_nr = next_spare++;
 696                 d = &sb->disks[rdev2->desc_nr];
 697                 nr_disks++;
 698                 d->number = rdev2->desc_nr;
 699                 d->major = MAJOR(rdev2->bdev->bd_dev);
 700                 d->minor = MINOR(rdev2->bdev->bd_dev);
 701                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
 702                         d->raid_disk = rdev2->raid_disk;
 703                 else
 704                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 705                 if (rdev2->faulty) {
 706                         d->state = (1<<MD_DISK_FAULTY);
 707                         failed++;
 708                 } else if (rdev2->in_sync) {
 709                         d->state = (1<<MD_DISK_ACTIVE);
 710                         d->state |= (1<<MD_DISK_SYNC);
 711                         active++;
 712                         working++;
 713                 } else {
 714                         d->state = 0;
 715                         spare++;
 716                         working++;
 717                 }
 718         }
 719
 720         /* now set the "removed" and "faulty" bits on any missing devices */
 721         for (i=0 ; i < mddev->raid_disks ; i++) {
 722                 mdp_disk_t *d = &sb->disks[i];
 723                 if (d->state == 0 && d->number == 0) {
 724                         d->number = i;
 725                         d->raid_disk = i;
 726                         d->state = (1<<MD_DISK_REMOVED);
 727                         d->state |= (1<<MD_DISK_FAULTY);
 728                         failed++;
 729                 }
 730         }
 731         sb->nr_disks = nr_disks;
 732         sb->active_disks = active;
 733         sb->working_disks = working;
 734         sb->failed_disks = failed;
 735         sb->spare_disks = spare;
 736
 737         sb->this_disk = sb->disks[rdev->desc_nr];
 738         sb->sb_csum = calc_sb_csum(sb);
 739 }
 740
 741 /*
 742  * version 1 superblock
 743  */
 744
 745 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 746 {
 747         unsigned int disk_csum, csum;
 748         int size = 256 + sb->max_dev*2;
 749
 750         disk_csum = sb->sb_csum;
 751         sb->sb_csum = 0;
 752         csum = csum_partial((void *)sb, size, 0);
 753         sb->sb_csum = disk_csum;
 754         return csum;
 755 }
 756
 757 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 758 {
 759         struct mdp_superblock_1 *sb;
 760         int ret;
 761         sector_t sb_offset;
 762         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 763
 764         /*
 765          * Calculate the position of the superblock.
 766          * It is always aligned to a 4K boundary and
 767          * depeding on minor_version, it can be:
 768          * 0: At least 8K, but less than 12K, from end of device
 769          * 1: At start of device
 770          * 2: 4K from start of device.
 771          */
 772         switch(minor_version) {
 773         case 0:
 774                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 775                 sb_offset -= 8*2;
 776                 sb_offset &= ~(4*2);
 777                 /* convert from sectors to K */
 778                 sb_offset /= 2;
 779                 break;
 780         case 1:
 781                 sb_offset = 0;
 782                 break;
 783         case 2:
 784                 sb_offset = 4;
 785                 break;
 786         default:
 787                 return -EINVAL;
 788         }
 789         rdev->sb_offset = sb_offset;
 790
 791         ret = read_disk_sb(rdev);
 792         if (ret) return ret;
 793
 794
 795         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 796
 797         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 798             sb->major_version != cpu_to_le32(1) ||
 799             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 800             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 801             sb->feature_map != 0)
 802                 return -EINVAL;
 803
 804         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 805                 printk("md: invalid superblock checksum on %s\n",
 806                         bdevname(rdev->bdev,b));
 807                 return -EINVAL;
 808         }
 809         rdev->preferred_minor = 0xffff;
 810         rdev->data_offset = le64_to_cpu(sb->data_offset);
 811
 812         if (refdev == 0)
 813                 return 1;
 814         else {
 815                 __u64 ev1, ev2;
 816                 struct mdp_superblock_1 *refsb =
 817                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 818
 819                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 820                     sb->level != refsb->level ||
 821                     sb->layout != refsb->layout ||
 822                     sb->chunksize != refsb->chunksize) {
 823                         printk(KERN_WARNING "md: %s has strangely different"
 824                                 " superblock to %s\n",
 825                                 bdevname(rdev->bdev,b),
 826                                 bdevname(refdev->bdev,b2));
 827                         return -EINVAL;
 828                 }
 829                 ev1 = le64_to_cpu(sb->events);
 830                 ev2 = le64_to_cpu(refsb->events);
 831
 832                 if (ev1 > ev2)
 833                         return 1;
 834         }
 835         if (minor_version)
 836                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 837         else
 838                 rdev->size = rdev->sb_offset;
 839         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 840                 return -EINVAL;
 841         rdev->size = le64_to_cpu(sb->data_size)/2;
 842         if (le32_to_cpu(sb->chunksize))
 843                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 844         return 0;
 845 }
 846
 847 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 848 {
 849         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 850
 851         if (mddev->raid_disks == 0) {
 852                 mddev->major_version = 1;
 853                 mddev->minor_version = 0;
 854                 mddev->patch_version = 0;
 855                 mddev->persistent = 1;
 856                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 857                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 858                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 859                 mddev->level = le32_to_cpu(sb->level);
 860                 mddev->layout = le32_to_cpu(sb->layout);
 861                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 862                 mddev->size = (u32)le64_to_cpu(sb->size);
 863                 mddev->events = le64_to_cpu(sb->events);
 864
 865                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 866                 memcpy(mddev->uuid, sb->set_uuid, 16);
 867
 868                 mddev->max_disks =  (4096-256)/2;
 869         } else {
 870                 __u64 ev1;
 871                 ev1 = le64_to_cpu(sb->events);
 872                 ++ev1;
 873                 if (ev1 < mddev->events)
 874                         return -EINVAL;
 875         }
 876
 877         if (mddev->level != LEVEL_MULTIPATH) {
 878                 int role;
 879                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 880                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 881                 switch(role) {
 882                 case 0xffff: /* spare */
 883                         rdev->in_sync = 0;
 884                         rdev->faulty = 0;
 885                         rdev->raid_disk = -1;
 886                         break;
 887                 case 0xfffe: /* faulty */
 888                         rdev->in_sync = 0;
 889                         rdev->faulty = 1;
 890                         rdev->raid_disk = -1;
 891                         break;
 892                 default:
 893                         rdev->in_sync = 1;
 894                         rdev->faulty = 0;
 895                         rdev->raid_disk = role;
 896                         break;
 897                 }
 898         }
 899         return 0;
 900 }
 901
 902 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 903 {
 904         struct mdp_superblock_1 *sb;
 905         struct list_head *tmp;
 906         mdk_rdev_t *rdev2;
 907         int max_dev, i;
 908         /* make rdev->sb match mddev and rdev data. */
 909
 910         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 911
 912         sb->feature_map = 0;
 913         sb->pad0 = 0;
 914         memset(sb->pad1, 0, sizeof(sb->pad1));
 915         memset(sb->pad2, 0, sizeof(sb->pad2));
 916         memset(sb->pad3, 0, sizeof(sb->pad3));
 917
 918         sb->utime = cpu_to_le64((__u64)mddev->utime);
 919         sb->events = cpu_to_le64(mddev->events);
 920         if (mddev->in_sync)
 921                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
 922         else
 923                 sb->resync_offset = cpu_to_le64(0);
 924
 925         max_dev = 0;
 926         ITERATE_RDEV(mddev,rdev2,tmp)
 927                 if (rdev2->desc_nr > max_dev)
 928                         max_dev = rdev2->desc_nr;
 929
 930         sb->max_dev = max_dev;
 931         for (i=0; i<max_dev;i++)
 932                 sb->dev_roles[max_dev] = cpu_to_le16(0xfffe);
 933
 934         ITERATE_RDEV(mddev,rdev2,tmp) {
 935                 i = rdev2->desc_nr;
 936                 if (rdev2->faulty)
 937                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
 938                 else if (rdev2->in_sync)
 939                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
 940                 else
 941                         sb->dev_roles[i] = cpu_to_le16(0xffff);
 942         }
 943
 944         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
 945 }
 946
 947
 948 struct super_type super_types[] = {
 949         [0] = {
 950                 .name   = "0.90.0",
 951                 .owner  = THIS_MODULE,
 952                 .load_super     = super_90_load,
 953                 .validate_super = super_90_validate,
 954                 .sync_super     = super_90_sync,
 955         },
 956         [1] = {
 957                 .name   = "md-1",
 958                 .owner  = THIS_MODULE,
 959                 .load_super     = super_1_load,
 960                 .validate_super = super_1_validate,
 961                 .sync_super     = super_1_sync,
 962         },
 963 };
 964
 965 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
 966 {
 967         struct list_head *tmp;
 968         mdk_rdev_t *rdev;
 969
 970         ITERATE_RDEV(mddev,rdev,tmp)
 971                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
 972                         return rdev;
 973
 974         return NULL;
 975 }
 976
 977 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
 978 {
 979         struct list_head *tmp;
 980         mdk_rdev_t *rdev;
 981
 982         ITERATE_RDEV(mddev1,rdev,tmp)
 983                 if (match_dev_unit(mddev2, rdev))
 984                         return 1;
 985
 986         return 0;
 987 }
 988
 989 static LIST_HEAD(pending_raid_disks);
 990
 991 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
 992 {
 993         mdk_rdev_t *same_pdev;
 994         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 995
 996         if (rdev->mddev) {
 997                 MD_BUG();
 998                 return -EINVAL;
 999         }
1000         same_pdev = match_dev_unit(mddev, rdev);
1001         if (same_pdev)
1002                 printk(KERN_WARNING
1003                         "%s: WARNING: %s appears to be on the same physical"
1004                         " disk as %s. True\n     protection against single-disk"
1005                         " failure might be compromised.\n",
1006                         mdname(mddev), bdevname(rdev->bdev,b),
1007                         bdevname(same_pdev->bdev,b2));
1008
1009         /* Verify rdev->desc_nr is unique.
1010          * If it is -1, assign a free number, else
1011          * check number is not in use
1012          */
1013         if (rdev->desc_nr < 0) {
1014                 int choice = 0;
1015                 if (mddev->pers) choice = mddev->raid_disks;
1016                 while (find_rdev_nr(mddev, choice))
1017                         choice++;
1018                 rdev->desc_nr = choice;
1019         } else {
1020                 if (find_rdev_nr(mddev, rdev->desc_nr))
1021                         return -EBUSY;
1022         }
1023
1024         list_add(&rdev->same_set, &mddev->disks);
1025         rdev->mddev = mddev;
1026         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1027         return 0;
1028 }
1029
1030 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1031 {
1032         char b[BDEVNAME_SIZE];
1033         if (!rdev->mddev) {
1034                 MD_BUG();
1035                 return;
1036         }
1037         list_del_init(&rdev->same_set);
1038         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1039         rdev->mddev = NULL;
1040 }
1041
1042 /*
1043  * prevent the device from being mounted, repartitioned or
1044  * otherwise reused by a RAID array (or any other kernel
1045  * subsystem), by opening the device. [simply getting an
1046  * inode is not enough, the SCSI module usage code needs
1047  * an explicit open() on the device]
1048  */
1049 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1050 {
1051         int err = 0;
1052         struct block_device *bdev;
1053
1054         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1055         if (IS_ERR(bdev))
1056                 return PTR_ERR(bdev);
1057         err = bd_claim(bdev, rdev);
1058         if (err) {
1059                 blkdev_put(bdev);
1060                 return err;
1061         }
1062         rdev->bdev = bdev;
1063         return err;
1064 }
1065
1066 static void unlock_rdev(mdk_rdev_t *rdev)
1067 {
1068         struct block_device *bdev = rdev->bdev;
1069         rdev->bdev = NULL;
1070         if (!bdev)
1071                 MD_BUG();
1072         bd_release(bdev);
1073         blkdev_put(bdev);
1074 }
1075
1076 void md_autodetect_dev(dev_t dev);
1077
1078 static void export_rdev(mdk_rdev_t * rdev)
1079 {
1080         char b[BDEVNAME_SIZE];
1081         printk(KERN_INFO "md: export_rdev(%s)\n",
1082                 bdevname(rdev->bdev,b));
1083         if (rdev->mddev)
1084                 MD_BUG();
1085         free_disk_sb(rdev);
1086         list_del_init(&rdev->same_set);
1087 #ifndef MODULE
1088         md_autodetect_dev(rdev->bdev->bd_dev);
1089 #endif
1090         unlock_rdev(rdev);
1091         kfree(rdev);
1092 }
1093
1094 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1095 {
1096         unbind_rdev_from_array(rdev);
1097         export_rdev(rdev);
1098 }
1099
1100 static void export_array(mddev_t *mddev)
1101 {
1102         struct list_head *tmp;
1103         mdk_rdev_t *rdev;
1104
1105         ITERATE_RDEV(mddev,rdev,tmp) {
1106                 if (!rdev->mddev) {
1107                         MD_BUG();
1108                         continue;
1109                 }
1110                 kick_rdev_from_array(rdev);
1111         }
1112         if (!list_empty(&mddev->disks))
1113                 MD_BUG();
1114         mddev->raid_disks = 0;
1115         mddev->major_version = 0;
1116 }
1117
1118 static void print_desc(mdp_disk_t *desc)
1119 {
1120         char b[BDEVNAME_SIZE];
1121
1122         printk(" DISK<N:%d,%s(%d,%d),R:%d,S:%d>\n", desc->number,
1123                 __bdevname(MKDEV(desc->major, desc->minor), b),
1124                 desc->major,desc->minor,desc->raid_disk,desc->state);
1125 }
1126
1127 static void print_sb(mdp_super_t *sb)
1128 {
1129         int i;
1130
1131         printk(KERN_INFO
1132                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1133                 sb->major_version, sb->minor_version, sb->patch_version,
1134                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1135                 sb->ctime);
1136         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1137                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1138                 sb->md_minor, sb->layout, sb->chunk_size);
1139         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1140                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1141                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1142                 sb->failed_disks, sb->spare_disks,
1143                 sb->sb_csum, (unsigned long)sb->events_lo);
1144
1145         printk(KERN_INFO);
1146         for (i = 0; i < MD_SB_DISKS; i++) {
1147                 mdp_disk_t *desc;
1148
1149                 desc = sb->disks + i;
1150                 if (desc->number || desc->major || desc->minor ||
1151                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1152                         printk("     D %2d: ", i);
1153                         print_desc(desc);
1154                 }
1155         }
1156         printk(KERN_INFO "md:     THIS: ");
1157         print_desc(&sb->this_disk);
1158
1159 }
1160
1161 static void print_rdev(mdk_rdev_t *rdev)
1162 {
1163         char b[BDEVNAME_SIZE];
1164         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1165                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1166                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1167         if (rdev->sb_loaded) {
1168                 printk(KERN_INFO "md: rdev superblock:\n");
1169                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1170         } else
1171                 printk(KERN_INFO "md: no rdev superblock!\n");
1172 }
1173
1174 void md_print_devices(void)
1175 {
1176         struct list_head *tmp, *tmp2;
1177         mdk_rdev_t *rdev;
1178         mddev_t *mddev;
1179         char b[BDEVNAME_SIZE];
1180
1181         printk("\n");
1182         printk("md:     **********************************\n");
1183         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1184         printk("md:     **********************************\n");
1185         ITERATE_MDDEV(mddev,tmp) {
1186                 printk("%s: ", mdname(mddev));
1187
1188                 ITERATE_RDEV(mddev,rdev,tmp2)
1189                         printk("<%s>", bdevname(rdev->bdev,b));
1190                 printk("\n");
1191
1192                 ITERATE_RDEV(mddev,rdev,tmp2)
1193                         print_rdev(rdev);
1194         }
1195         printk("md:     **********************************\n");
1196         printk("\n");
1197 }
1198
1199
1200 static int write_disk_sb(mdk_rdev_t * rdev)
1201 {
1202         char b[BDEVNAME_SIZE];
1203         if (!rdev->sb_loaded) {
1204                 MD_BUG();
1205                 return 1;
1206         }
1207         if (rdev->faulty) {
1208                 MD_BUG();
1209                 return 1;
1210         }
1211
1212         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1213                 bdevname(rdev->bdev,b),
1214                (unsigned long long)rdev->sb_offset);
1215
1216         if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE))
1217                 return 0;
1218
1219         printk("md: write_disk_sb failed for device %s\n",
1220                 bdevname(rdev->bdev,b));
1221         return 1;
1222 }
1223
1224 static void sync_sbs(mddev_t * mddev)
1225 {
1226         mdk_rdev_t *rdev;
1227         struct list_head *tmp;
1228
1229         ITERATE_RDEV(mddev,rdev,tmp) {
1230                 super_types[mddev->major_version].
1231                         sync_super(mddev, rdev);
1232                 rdev->sb_loaded = 1;
1233         }
1234 }
1235
1236 static void md_update_sb(mddev_t * mddev)
1237 {
1238         int err, count = 100;
1239         struct list_head *tmp;
1240         mdk_rdev_t *rdev;
1241
1242         mddev->sb_dirty = 0;
1243 repeat:
1244         mddev->utime = get_seconds();
1245         mddev->events ++;
1246
1247         if (!mddev->events) {
1248                 /*
1249                  * oops, this 64-bit counter should never wrap.
1250                  * Either we are in around ~1 trillion A.C., assuming
1251                  * 1 reboot per second, or we have a bug:
1252                  */
1253                 MD_BUG();
1254                 mddev->events --;
1255         }
1256         sync_sbs(mddev);
1257
1258         /*
1259          * do not write anything to disk if using
1260          * nonpersistent superblocks
1261          */
1262         if (!mddev->persistent)
1263                 return;
1264
1265         dprintk(KERN_INFO
1266                 "md: updating %s RAID superblock on device (in sync %d)\n",
1267                 mdname(mddev),mddev->in_sync);
1268
1269         err = 0;
1270         ITERATE_RDEV(mddev,rdev,tmp) {
1271                 char b[BDEVNAME_SIZE];
1272                 dprintk(KERN_INFO "md: ");
1273                 if (rdev->faulty)
1274                         dprintk("(skipping faulty ");
1275
1276                 dprintk("%s ", bdevname(rdev->bdev,b));
1277                 if (!rdev->faulty) {
1278                         err += write_disk_sb(rdev);
1279                 } else
1280                         dprintk(")\n");
1281                 if (!err && mddev->level == LEVEL_MULTIPATH)
1282                         /* only need to write one superblock... */
1283                         break;
1284         }
1285         if (err) {
1286                 if (--count) {
1287                         printk(KERN_ERR "md: errors occurred during superblock"
1288                                 " update, repeating\n");
1289                         goto repeat;
1290                 }
1291                 printk(KERN_ERR \
1292                         "md: excessive errors occurred during superblock update, exiting\n");
1293         }
1294 }
1295
1296 /*
1297  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1298  *
1299  * mark the device faulty if:
1300  *
1301  *   - the device is nonexistent (zero size)
1302  *   - the device has no valid superblock
1303  *
1304  * a faulty rdev _never_ has rdev->sb set.
1305  */
1306 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1307 {
1308         char b[BDEVNAME_SIZE];
1309         int err;
1310         mdk_rdev_t *rdev;
1311         sector_t size;
1312
1313         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1314         if (!rdev) {
1315                 printk(KERN_ERR "md: could not alloc mem for %s!\n",
1316                         __bdevname(newdev, b));
1317                 return ERR_PTR(-ENOMEM);
1318         }
1319         memset(rdev, 0, sizeof(*rdev));
1320
1321         if ((err = alloc_disk_sb(rdev)))
1322                 goto abort_free;
1323
1324         err = lock_rdev(rdev, newdev);
1325         if (err) {
1326                 printk(KERN_ERR "md: could not lock %s.\n",
1327                         __bdevname(newdev, b));
1328                 goto abort_free;
1329         }
1330         rdev->desc_nr = -1;
1331         rdev->faulty = 0;
1332         rdev->in_sync = 0;
1333         rdev->data_offset = 0;
1334         atomic_set(&rdev->nr_pending, 0);
1335
1336         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1337         if (!size) {
1338                 printk(KERN_WARNING
1339                         "md: %s has zero or unknown size, marking faulty!\n",
1340                         bdevname(rdev->bdev,b));
1341                 err = -EINVAL;
1342                 goto abort_free;
1343         }
1344
1345         if (super_format >= 0) {
1346                 err = super_types[super_format].
1347                         load_super(rdev, NULL, super_minor);
1348                 if (err == -EINVAL) {
1349                         printk(KERN_WARNING
1350                                 "md: %s has invalid sb, not importing!\n",
1351                                 bdevname(rdev->bdev,b));
1352                         goto abort_free;
1353                 }
1354                 if (err < 0) {
1355                         printk(KERN_WARNING
1356                                 "md: could not read %s's sb, not importing!\n",
1357                                 bdevname(rdev->bdev,b));
1358                         goto abort_free;
1359                 }
1360         }
1361         INIT_LIST_HEAD(&rdev->same_set);
1362
1363         return rdev;
1364
1365 abort_free:
1366         if (rdev->sb_page) {
1367                 if (rdev->bdev)
1368                         unlock_rdev(rdev);
1369                 free_disk_sb(rdev);
1370         }
1371         kfree(rdev);
1372         return ERR_PTR(err);
1373 }
1374
1375 /*
1376  * Check a full RAID array for plausibility
1377  */
1378
1379
1380 static int analyze_sbs(mddev_t * mddev)
1381 {
1382         int i;
1383         struct list_head *tmp;
1384         mdk_rdev_t *rdev, *freshest;
1385         char b[BDEVNAME_SIZE];
1386
1387         freshest = NULL;
1388         ITERATE_RDEV(mddev,rdev,tmp)
1389                 switch (super_types[mddev->major_version].
1390                         load_super(rdev, freshest, mddev->minor_version)) {
1391                 case 1:
1392                         freshest = rdev;
1393                         break;
1394                 case 0:
1395                         break;
1396                 default:
1397                         printk( KERN_ERR \
1398                                 "md: fatal superblock inconsistency in %s"
1399                                 " -- removing from array\n",
1400                                 bdevname(rdev->bdev,b));
1401                         kick_rdev_from_array(rdev);
1402                 }
1403
1404
1405         super_types[mddev->major_version].
1406                 validate_super(mddev, freshest);
1407
1408         i = 0;
1409         ITERATE_RDEV(mddev,rdev,tmp) {
1410                 if (rdev != freshest)
1411                         if (super_types[mddev->major_version].
1412                             validate_super(mddev, rdev)) {
1413                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1414                                         " from array!\n",
1415                                         bdevname(rdev->bdev,b));
1416                                 kick_rdev_from_array(rdev);
1417                                 continue;
1418                         }
1419                 if (mddev->level == LEVEL_MULTIPATH) {
1420                         rdev->desc_nr = i++;
1421                         rdev->raid_disk = rdev->desc_nr;
1422                         rdev->in_sync = 1;
1423                 }
1424         }
1425
1426
1427         /*
1428          * Check if we can support this RAID array
1429          */
1430         if (mddev->major_version != MD_MAJOR_VERSION ||
1431                         mddev->minor_version > MD_MINOR_VERSION) {
1432                 printk(KERN_ALERT
1433                         "md: %s: unsupported raid array version %d.%d.%d\n",
1434                         mdname(mddev), mddev->major_version,
1435                         mddev->minor_version, mddev->patch_version);
1436                 goto abort;
1437         }
1438
1439         if ((mddev->recovery_cp != MaxSector) &&
1440             ((mddev->level == 1) ||
1441              ((mddev->level >= 4) && (mddev->level <= 6))))
1442                 printk(KERN_ERR "md: %s: raid array is not clean"
1443                        " -- starting background reconstruction\n",
1444                        mdname(mddev));
1445
1446         return 0;
1447 abort:
1448         return 1;
1449 }
1450
1451 int mdp_major = 0;
1452
1453 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1454 {
1455         static DECLARE_MUTEX(disks_sem);
1456         mddev_t *mddev = mddev_find(dev);
1457         struct gendisk *disk;
1458         int partitioned = (MAJOR(dev) != MD_MAJOR);
1459         int shift = partitioned ? MdpMinorShift : 0;
1460         int unit = MINOR(dev) >> shift;
1461
1462         if (!mddev)
1463                 return NULL;
1464
1465         down(&disks_sem);
1466         if (mddev->gendisk) {
1467                 up(&disks_sem);
1468                 mddev_put(mddev);
1469                 return NULL;
1470         }
1471         disk = alloc_disk(1 << shift);
1472         if (!disk) {
1473                 up(&disks_sem);
1474                 mddev_put(mddev);
1475                 return NULL;
1476         }
1477         disk->major = MAJOR(dev);
1478         disk->first_minor = unit << shift;
1479         if (partitioned)
1480                 sprintf(disk->disk_name, "md_d%d", unit);
1481         else
1482                 sprintf(disk->disk_name, "md%d", unit);
1483         disk->fops = &md_fops;
1484         disk->private_data = mddev;
1485         disk->queue = mddev->queue;
1486         add_disk(disk);
1487         mddev->gendisk = disk;
1488         up(&disks_sem);
1489         return NULL;
1490 }
1491
1492 void md_wakeup_thread(mdk_thread_t *thread);
1493
1494 static void md_safemode_timeout(unsigned long data)
1495 {
1496         mddev_t *mddev = (mddev_t *) data;
1497
1498         mddev->safemode = 1;
1499         md_wakeup_thread(mddev->thread);
1500 }
1501
1502
1503 static int do_md_run(mddev_t * mddev)
1504 {
1505         int pnum, err;
1506         int chunk_size;
1507         struct list_head *tmp;
1508         mdk_rdev_t *rdev;
1509         struct gendisk *disk;
1510         char b[BDEVNAME_SIZE];
1511
1512         if (list_empty(&mddev->disks)) {
1513                 MD_BUG();
1514                 return -EINVAL;
1515         }
1516
1517         if (mddev->pers)
1518                 return -EBUSY;
1519
1520         /*
1521          * Analyze all RAID superblock(s)
1522          */
1523         if (!mddev->raid_disks && analyze_sbs(mddev)) {
1524                 MD_BUG();
1525                 return -EINVAL;
1526         }
1527
1528         chunk_size = mddev->chunk_size;
1529         pnum = level_to_pers(mddev->level);
1530
1531         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1532                 if (!chunk_size) {
1533                         /*
1534                          * 'default chunksize' in the old md code used to
1535                          * be PAGE_SIZE, baaad.
1536                          * we abort here to be on the safe side. We don't
1537                          * want to continue the bad practice.
1538                          */
1539                         printk(KERN_ERR
1540                                 "no chunksize specified, see 'man raidtab'\n");
1541                         return -EINVAL;
1542                 }
1543                 if (chunk_size > MAX_CHUNK_SIZE) {
1544                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1545                                 chunk_size, MAX_CHUNK_SIZE);
1546                         return -EINVAL;
1547                 }
1548                 /*
1549                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1550                  */
1551                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1552                         MD_BUG();
1553                         return -EINVAL;
1554                 }
1555                 if (chunk_size < PAGE_SIZE) {
1556                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1557                                 chunk_size, PAGE_SIZE);
1558                         return -EINVAL;
1559                 }
1560
1561                 /* devices must have minimum size of one chunk */
1562                 ITERATE_RDEV(mddev,rdev,tmp) {
1563                         if (rdev->faulty)
1564                                 continue;
1565                         if (rdev->size < chunk_size / 1024) {
1566                                 printk(KERN_WARNING
1567                                         "md: Dev %s smaller than chunk_size:"
1568                                         " %lluk < %dk\n",
1569                                         bdevname(rdev->bdev,b),
1570                                         (unsigned long long)rdev->size,
1571                                         chunk_size / 1024);
1572                                 return -EINVAL;
1573                         }
1574                 }
1575         }
1576
1577         if (pnum >= MAX_PERSONALITY) {
1578                 MD_BUG();
1579                 return -EINVAL;
1580         }
1581
1582 #ifdef CONFIG_KMOD
1583         if (!pers[pnum])
1584         {
1585                 request_module("md-personality-%d", pnum);
1586         }
1587 #endif
1588
1589         /*
1590          * Drop all container device buffers, from now on
1591          * the only valid external interface is through the md
1592          * device.
1593          * Also find largest hardsector size
1594          */
1595         ITERATE_RDEV(mddev,rdev,tmp) {
1596                 if (rdev->faulty)
1597                         continue;
1598                 sync_blockdev(rdev->bdev);
1599                 invalidate_bdev(rdev->bdev, 0);
1600         }
1601
1602         md_probe(mddev->unit, NULL, NULL);
1603         disk = mddev->gendisk;
1604         if (!disk)
1605                 return -ENOMEM;
1606
1607         spin_lock(&pers_lock);
1608         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1609                 spin_unlock(&pers_lock);
1610                 printk(KERN_ERR "md: personality %d is not loaded!\n",
1611                        pnum);
1612                 return -EINVAL;
1613         }
1614
1615         mddev->pers = pers[pnum];
1616         spin_unlock(&pers_lock);
1617
1618         err = mddev->pers->run(mddev);
1619         if (err) {
1620                 printk(KERN_ERR "md: pers->run() failed ...\n");
1621                 module_put(mddev->pers->owner);
1622                 mddev->pers = NULL;
1623                 return -EINVAL;
1624         }
1625         atomic_set(&mddev->writes_pending,0);
1626         mddev->safemode = 0;
1627         mddev->safemode_timer.function = md_safemode_timeout;
1628         mddev->safemode_timer.data = (unsigned long) mddev;
1629         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1630         mddev->in_sync = 1;
1631
1632         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1633
1634         if (mddev->sb_dirty)
1635                 md_update_sb(mddev);
1636
1637         set_capacity(disk, mddev->array_size<<1);
1638
1639         /* If we call blk_queue_make_request here, it will
1640          * re-initialise max_sectors etc which may have been
1641          * refined inside -> run.  So just set the bits we need to set.
1642          * Most initialisation happended when we called
1643          * blk_queue_make_request(..., md_fail_request)
1644          * earlier.
1645          */
1646         mddev->queue->queuedata = mddev;
1647         mddev->queue->make_request_fn = mddev->pers->make_request;
1648
1649         mddev->changed = 1;
1650         return 0;
1651 }
1652
1653 static int restart_array(mddev_t *mddev)
1654 {
1655         struct gendisk *disk = mddev->gendisk;
1656         int err;
1657
1658         /*
1659          * Complain if it has no devices
1660          */
1661         err = -ENXIO;
1662         if (list_empty(&mddev->disks))
1663                 goto out;
1664
1665         if (mddev->pers) {
1666                 err = -EBUSY;
1667                 if (!mddev->ro)
1668                         goto out;
1669
1670                 mddev->safemode = 0;
1671                 mddev->ro = 0;
1672                 set_disk_ro(disk, 0);
1673
1674                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1675                         mdname(mddev));
1676                 /*
1677                  * Kick recovery or resync if necessary
1678                  */
1679                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1680                 md_wakeup_thread(mddev->thread);
1681                 err = 0;
1682         } else {
1683                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1684                         mdname(mddev));
1685                 err = -EINVAL;
1686         }
1687
1688 out:
1689         return err;
1690 }
1691
1692 static int do_md_stop(mddev_t * mddev, int ro)
1693 {
1694         int err = 0;
1695         struct gendisk *disk = mddev->gendisk;
1696
1697         if (mddev->pers) {
1698                 if (atomic_read(&mddev->active)>2) {
1699                         printk("md: %s still in use.\n",mdname(mddev));
1700                         return -EBUSY;
1701                 }
1702
1703                 if (mddev->sync_thread) {
1704                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1705                         md_unregister_thread(mddev->sync_thread);
1706                         mddev->sync_thread = NULL;
1707                 }
1708
1709                 del_timer_sync(&mddev->safemode_timer);
1710
1711                 invalidate_partition(disk, 0);
1712
1713                 if (ro) {
1714                         err  = -ENXIO;
1715                         if (mddev->ro)
1716                                 goto out;
1717                         mddev->ro = 1;
1718                 } else {
1719                         if (mddev->ro)
1720                                 set_disk_ro(disk, 0);
1721                         blk_queue_make_request(mddev->queue, md_fail_request);
1722                         mddev->pers->stop(mddev);
1723                         module_put(mddev->pers->owner);
1724                         mddev->pers = NULL;
1725                         if (mddev->ro)
1726                                 mddev->ro = 0;
1727                 }
1728                 if (!mddev->in_sync) {
1729                         /* mark array as shutdown cleanly */
1730                         mddev->in_sync = 1;
1731                         md_update_sb(mddev);
1732                 }
1733                 if (ro)
1734                         set_disk_ro(disk, 1);
1735         }
1736         /*
1737          * Free resources if final stop
1738          */
1739         if (!ro) {
1740                 struct gendisk *disk;
1741                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1742
1743                 export_array(mddev);
1744
1745                 mddev->array_size = 0;
1746                 disk = mddev->gendisk;
1747                 if (disk)
1748                         set_capacity(disk, 0);
1749                 mddev->changed = 1;
1750         } else
1751                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1752                         mdname(mddev));
1753         err = 0;
1754 out:
1755         return err;
1756 }
1757
1758 static void autorun_array(mddev_t *mddev)
1759 {
1760         mdk_rdev_t *rdev;
1761         struct list_head *tmp;
1762         int err;
1763
1764         if (list_empty(&mddev->disks)) {
1765                 MD_BUG();
1766                 return;
1767         }
1768
1769         printk(KERN_INFO "md: running: ");
1770
1771         ITERATE_RDEV(mddev,rdev,tmp) {
1772                 char b[BDEVNAME_SIZE];
1773                 printk("<%s>", bdevname(rdev->bdev,b));
1774         }
1775         printk("\n");
1776
1777         err = do_md_run (mddev);
1778         if (err) {
1779                 printk(KERN_WARNING "md :do_md_run() returned %d\n", err);
1780                 do_md_stop (mddev, 0);
1781         }
1782 }
1783
1784 /*
1785  * lets try to run arrays based on all disks that have arrived
1786  * until now. (those are in pending_raid_disks)
1787  *
1788  * the method: pick the first pending disk, collect all disks with
1789  * the same UUID, remove all from the pending list and put them into
1790  * the 'same_array' list. Then order this list based on superblock
1791  * update time (freshest comes first), kick out 'old' disks and
1792  * compare superblocks. If everything's fine then run it.
1793  *
1794  * If "unit" is allocated, then bump its reference count
1795  */
1796 static void autorun_devices(int part)
1797 {
1798         struct list_head candidates;
1799         struct list_head *tmp;
1800         mdk_rdev_t *rdev0, *rdev;
1801         mddev_t *mddev;
1802         char b[BDEVNAME_SIZE];
1803
1804         printk(KERN_INFO "md: autorun ...\n");
1805         while (!list_empty(&pending_raid_disks)) {
1806                 dev_t dev;
1807                 rdev0 = list_entry(pending_raid_disks.next,
1808                                          mdk_rdev_t, same_set);
1809
1810                 printk(KERN_INFO "md: considering %s ...\n",
1811                         bdevname(rdev0->bdev,b));
1812                 INIT_LIST_HEAD(&candidates);
1813                 ITERATE_RDEV_PENDING(rdev,tmp)
1814                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1815                                 printk(KERN_INFO "md:  adding %s ...\n",
1816                                         bdevname(rdev->bdev,b));
1817                                 list_move(&rdev->same_set, &candidates);
1818                         }
1819                 /*
1820                  * now we have a set of devices, with all of them having
1821                  * mostly sane superblocks. It's time to allocate the
1822                  * mddev.
1823                  */
1824                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1825                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1826                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1827                         break;
1828                 }
1829                 if (part)
1830                         dev = MKDEV(mdp_major,
1831                                     rdev0->preferred_minor << MdpMinorShift);
1832                 else
1833                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1834
1835                 md_probe(dev, NULL, NULL);
1836                 mddev = mddev_find(dev);
1837                 if (!mddev) {
1838                         printk(KERN_ERR
1839                                 "md: cannot allocate memory for md drive.\n");
1840                         break;
1841                 }
1842                 if (mddev_lock(mddev))
1843                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1844                                mdname(mddev));
1845                 else if (mddev->raid_disks || mddev->major_version
1846                          || !list_empty(&mddev->disks)) {
1847                         printk(KERN_WARNING
1848                                 "md: %s already running, cannot run %s\n",
1849                                 mdname(mddev), bdevname(rdev0->bdev,b));
1850                         mddev_unlock(mddev);
1851                 } else {
1852                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1853                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1854                                 list_del_init(&rdev->same_set);
1855                                 if (bind_rdev_to_array(rdev, mddev))
1856                                         export_rdev(rdev);
1857                         }
1858                         autorun_array(mddev);
1859                         mddev_unlock(mddev);
1860                 }
1861                 /* on success, candidates will be empty, on error
1862                  * it won't...
1863                  */
1864                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1865                         export_rdev(rdev);
1866                 mddev_put(mddev);
1867         }
1868         printk(KERN_INFO "md: ... autorun DONE.\n");
1869 }
1870
1871 /*
1872  * import RAID devices based on one partition
1873  * if possible, the array gets run as well.
1874  */
1875
1876 static int autostart_array(dev_t startdev)
1877 {
1878         char b[BDEVNAME_SIZE];
1879         int err = -EINVAL, i;
1880         mdp_super_t *sb = NULL;
1881         mdk_rdev_t *start_rdev = NULL, *rdev;
1882
1883         start_rdev = md_import_device(startdev, 0, 0);
1884         if (IS_ERR(start_rdev)) {
1885                 printk(KERN_WARNING "md: could not import %s!\n",
1886                         __bdevname(startdev, b));
1887                 return err;
1888         }
1889
1890         /* NOTE: this can only work for 0.90.0 superblocks */
1891         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1892         if (sb->major_version != 0 ||
1893             sb->minor_version != 90 ) {
1894                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1895                 export_rdev(start_rdev);
1896                 return err;
1897         }
1898
1899         if (start_rdev->faulty) {
1900                 printk(KERN_WARNING
1901                         "md: can not autostart based on faulty %s!\n",
1902                         bdevname(start_rdev->bdev,b));
1903                 export_rdev(start_rdev);
1904                 return err;
1905         }
1906         list_add(&start_rdev->same_set, &pending_raid_disks);
1907
1908         for (i = 0; i < MD_SB_DISKS; i++) {
1909                 mdp_disk_t *desc = sb->disks + i;
1910                 dev_t dev = MKDEV(desc->major, desc->minor);
1911
1912                 if (!dev)
1913                         continue;
1914                 if (dev == startdev)
1915                         continue;
1916                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
1917                         continue;
1918                 rdev = md_import_device(dev, 0, 0);
1919                 if (IS_ERR(rdev)) {
1920                         printk(KERN_WARNING "md: could not import %s,"
1921                                 " trying to run array nevertheless.\n",
1922                                 __bdevname(dev, b));
1923                         continue;
1924                 }
1925                 list_add(&rdev->same_set, &pending_raid_disks);
1926         }
1927
1928         /*
1929          * possibly return codes
1930          */
1931         autorun_devices(0);
1932         return 0;
1933
1934 }
1935
1936
1937 static int get_version(void * arg)
1938 {
1939         mdu_version_t ver;
1940
1941         ver.major = MD_MAJOR_VERSION;
1942         ver.minor = MD_MINOR_VERSION;
1943         ver.patchlevel = MD_PATCHLEVEL_VERSION;
1944
1945         if (copy_to_user(arg, &ver, sizeof(ver)))
1946                 return -EFAULT;
1947
1948         return 0;
1949 }
1950
1951 static int get_array_info(mddev_t * mddev, void * arg)
1952 {
1953         mdu_array_info_t info;
1954         int nr,working,active,failed,spare;
1955         mdk_rdev_t *rdev;
1956         struct list_head *tmp;
1957
1958         nr=working=active=failed=spare=0;
1959         ITERATE_RDEV(mddev,rdev,tmp) {
1960                 nr++;
1961                 if (rdev->faulty)
1962                         failed++;
1963                 else {
1964                         working++;
1965                         if (rdev->in_sync)
1966                                 active++;
1967                         else
1968                                 spare++;
1969                 }
1970         }
1971
1972         info.major_version = mddev->major_version;
1973         info.minor_version = mddev->minor_version;
1974         info.patch_version = 1;
1975         info.ctime         = mddev->ctime;
1976         info.level         = mddev->level;
1977         info.size          = mddev->size;
1978         info.nr_disks      = nr;
1979         info.raid_disks    = mddev->raid_disks;
1980         info.md_minor      = mddev->md_minor;
1981         info.not_persistent= !mddev->persistent;
1982
1983         info.utime         = mddev->utime;
1984         info.state         = 0;
1985         if (mddev->in_sync)
1986                 info.state = (1<<MD_SB_CLEAN);
1987         info.active_disks  = active;
1988         info.working_disks = working;
1989         info.failed_disks  = failed;
1990         info.spare_disks   = spare;
1991
1992         info.layout        = mddev->layout;
1993         info.chunk_size    = mddev->chunk_size;
1994
1995         if (copy_to_user(arg, &info, sizeof(info)))
1996                 return -EFAULT;
1997
1998         return 0;
1999 }
2000
2001 static int get_disk_info(mddev_t * mddev, void * arg)
2002 {
2003         mdu_disk_info_t info;
2004         unsigned int nr;
2005         mdk_rdev_t *rdev;
2006
2007         if (copy_from_user(&info, arg, sizeof(info)))
2008                 return -EFAULT;
2009
2010         nr = info.number;
2011
2012         rdev = find_rdev_nr(mddev, nr);
2013         if (rdev) {
2014                 info.major = MAJOR(rdev->bdev->bd_dev);
2015                 info.minor = MINOR(rdev->bdev->bd_dev);
2016                 info.raid_disk = rdev->raid_disk;
2017                 info.state = 0;
2018                 if (rdev->faulty)
2019                         info.state |= (1<<MD_DISK_FAULTY);
2020                 else if (rdev->in_sync) {
2021                         info.state |= (1<<MD_DISK_ACTIVE);
2022                         info.state |= (1<<MD_DISK_SYNC);
2023                 }
2024         } else {
2025                 info.major = info.minor = 0;
2026                 info.raid_disk = -1;
2027                 info.state = (1<<MD_DISK_REMOVED);
2028         }
2029
2030         if (copy_to_user(arg, &info, sizeof(info)))
2031                 return -EFAULT;
2032
2033         return 0;
2034 }
2035
2036 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2037 {
2038         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2039         mdk_rdev_t *rdev;
2040         dev_t dev = MKDEV(info->major,info->minor);
2041
2042         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2043                 return -EOVERFLOW;
2044
2045         if (!mddev->raid_disks) {
2046                 int err;
2047                 /* expecting a device which has a superblock */
2048                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2049                 if (IS_ERR(rdev)) {
2050                         printk(KERN_WARNING
2051                                 "md: md_import_device returned %ld\n",
2052                                 PTR_ERR(rdev));
2053                         return PTR_ERR(rdev);
2054                 }
2055                 if (!list_empty(&mddev->disks)) {
2056                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2057                                                         mdk_rdev_t, same_set);
2058                         int err = super_types[mddev->major_version]
2059                                 .load_super(rdev, rdev0, mddev->minor_version);
2060                         if (err < 0) {
2061                                 printk(KERN_WARNING
2062                                         "md: %s has different UUID to %s\n",
2063                                         bdevname(rdev->bdev,b),
2064                                         bdevname(rdev0->bdev,b2));
2065                                 export_rdev(rdev);
2066                                 return -EINVAL;
2067                         }
2068                 }
2069                 err = bind_rdev_to_array(rdev, mddev);
2070                 if (err)
2071                         export_rdev(rdev);
2072                 return err;
2073         }
2074
2075         /*
2076          * add_new_disk can be used once the array is assembled
2077          * to add "hot spares".  They must already have a superblock
2078          * written
2079          */
2080         if (mddev->pers) {
2081                 int err;
2082                 if (!mddev->pers->hot_add_disk) {
2083                         printk(KERN_WARNING
2084                                 "%s: personality does not support diskops!\n",
2085                                mdname(mddev));
2086                         return -EINVAL;
2087                 }
2088                 rdev = md_import_device(dev, mddev->major_version,
2089                                         mddev->minor_version);
2090                 if (IS_ERR(rdev)) {
2091                         printk(KERN_WARNING
2092                                 "md: md_import_device returned %ld\n",
2093                                 PTR_ERR(rdev));
2094                         return PTR_ERR(rdev);
2095                 }
2096                 rdev->in_sync = 0; /* just to be sure */
2097                 rdev->raid_disk = -1;
2098                 err = bind_rdev_to_array(rdev, mddev);
2099                 if (err)
2100                         export_rdev(rdev);
2101                 if (mddev->thread)
2102                         md_wakeup_thread(mddev->thread);
2103                 return err;
2104         }
2105
2106         /* otherwise, add_new_disk is only allowed
2107          * for major_version==0 superblocks
2108          */
2109         if (mddev->major_version != 0) {
2110                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2111                        mdname(mddev));
2112                 return -EINVAL;
2113         }
2114
2115         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2116                 int err;
2117                 rdev = md_import_device (dev, -1, 0);
2118                 if (IS_ERR(rdev)) {
2119                         printk(KERN_WARNING
2120                                 "md: error, md_import_device() returned %ld\n",
2121                                 PTR_ERR(rdev));
2122                         return PTR_ERR(rdev);
2123                 }
2124                 rdev->desc_nr = info->number;
2125                 if (info->raid_disk < mddev->raid_disks)
2126                         rdev->raid_disk = info->raid_disk;
2127                 else
2128                         rdev->raid_disk = -1;
2129
2130                 rdev->faulty = 0;
2131                 if (rdev->raid_disk < mddev->raid_disks)
2132                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2133                 else
2134                         rdev->in_sync = 0;
2135
2136                 err = bind_rdev_to_array(rdev, mddev);
2137                 if (err) {
2138                         export_rdev(rdev);
2139                         return err;
2140                 }
2141
2142                 if (!mddev->persistent) {
2143                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2144                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2145                 } else
2146                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2147                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2148
2149                 if (!mddev->size || (mddev->size > rdev->size))
2150                         mddev->size = rdev->size;
2151         }
2152
2153         return 0;
2154 }
2155
2156 static int hot_generate_error(mddev_t * mddev, dev_t dev)
2157 {
2158         char b[BDEVNAME_SIZE];
2159         struct request_queue *q;
2160         mdk_rdev_t *rdev;
2161
2162         if (!mddev->pers)
2163                 return -ENODEV;
2164
2165         printk(KERN_INFO "md: trying to generate %s error in %s ... \n",
2166                 __bdevname(dev, b), mdname(mddev));
2167
2168         rdev = find_rdev(mddev, dev);
2169         if (!rdev) {
2170                 /* MD_BUG(); */ /* like hell - it's not a driver bug */
2171                 return -ENXIO;
2172         }
2173
2174         if (rdev->desc_nr == -1) {
2175                 MD_BUG();
2176                 return -EINVAL;
2177         }
2178         if (!rdev->in_sync)
2179                 return -ENODEV;
2180
2181         q = bdev_get_queue(rdev->bdev);
2182         if (!q) {
2183                 MD_BUG();
2184                 return -ENODEV;
2185         }
2186         printk(KERN_INFO "md: okay, generating error!\n");
2187 //      q->oneshot_error = 1; // disabled for now
2188
2189         return 0;
2190 }
2191
2192 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2193 {
2194         char b[BDEVNAME_SIZE];
2195         mdk_rdev_t *rdev;
2196
2197         if (!mddev->pers)
2198                 return -ENODEV;
2199
2200         printk(KERN_INFO "md: trying to remove %s from %s ... \n",
2201                 __bdevname(dev, b), mdname(mddev));
2202
2203         rdev = find_rdev(mddev, dev);
2204         if (!rdev)
2205                 return -ENXIO;
2206
2207         if (rdev->raid_disk >= 0)
2208                 goto busy;
2209
2210         kick_rdev_from_array(rdev);
2211         md_update_sb(mddev);
2212
2213         return 0;
2214 busy:
2215         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2216                 bdevname(rdev->bdev,b), mdname(mddev));
2217         return -EBUSY;
2218 }
2219
2220 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2221 {
2222         char b[BDEVNAME_SIZE];
2223         int err;
2224         unsigned int size;
2225         mdk_rdev_t *rdev;
2226
2227         if (!mddev->pers)
2228                 return -ENODEV;
2229
2230         printk(KERN_INFO "md: trying to hot-add %s to %s ... \n",
2231                 __bdevname(dev, b), mdname(mddev));
2232
2233         if (mddev->major_version != 0) {
2234                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2235                         " version-0 superblocks.\n",
2236                         mdname(mddev));
2237                 return -EINVAL;
2238         }
2239         if (!mddev->pers->hot_add_disk) {
2240                 printk(KERN_WARNING
2241                         "%s: personality does not support diskops!\n",
2242                         mdname(mddev));
2243                 return -EINVAL;
2244         }
2245
2246         rdev = md_import_device (dev, -1, 0);
2247         if (IS_ERR(rdev)) {
2248                 printk(KERN_WARNING
2249                         "md: error, md_import_device() returned %ld\n",
2250                         PTR_ERR(rdev));
2251                 return -EINVAL;
2252         }
2253
2254         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2255         size = calc_dev_size(rdev, mddev->chunk_size);
2256         rdev->size = size;
2257
2258         if (size < mddev->size) {
2259                 printk(KERN_WARNING
2260                         "%s: disk size %llu blocks < array size %llu\n",
2261                         mdname(mddev), (unsigned long long)size,
2262                         (unsigned long long)mddev->size);
2263                 err = -ENOSPC;
2264                 goto abort_export;
2265         }
2266
2267         if (rdev->faulty) {
2268                 printk(KERN_WARNING
2269                         "md: can not hot-add faulty %s disk to %s!\n",
2270                         bdevname(rdev->bdev,b), mdname(mddev));
2271                 err = -EINVAL;
2272                 goto abort_export;
2273         }
2274         rdev->in_sync = 0;
2275         rdev->desc_nr = -1;
2276         bind_rdev_to_array(rdev, mddev);
2277
2278         /*
2279          * The rest should better be atomic, we can have disk failures
2280          * noticed in interrupt contexts ...
2281          */
2282
2283         if (rdev->desc_nr == mddev->max_disks) {
2284                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2285                         mdname(mddev));
2286                 err = -EBUSY;
2287                 goto abort_unbind_export;
2288         }
2289
2290         rdev->raid_disk = -1;
2291
2292         md_update_sb(mddev);
2293
2294         /*
2295          * Kick recovery, maybe this spare has to be added to the
2296          * array immediately.
2297          */
2298         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2299         md_wakeup_thread(mddev->thread);
2300
2301         return 0;
2302
2303 abort_unbind_export:
2304         unbind_rdev_from_array(rdev);
2305
2306 abort_export:
2307         export_rdev(rdev);
2308         return err;
2309 }
2310
2311 /*
2312  * set_array_info is used two different ways
2313  * The original usage is when creating a new array.
2314  * In this usage, raid_disks is > 0 and it together with
2315  *  level, size, not_persistent,layout,chunksize determine the
2316  *  shape of the array.
2317  *  This will always create an array with a type-0.90.0 superblock.
2318  * The newer usage is when assembling an array.
2319  *  In this case raid_disks will be 0, and the major_version field is
2320  *  use to determine which style super-blocks are to be found on the devices.
2321  *  The minor and patch _version numbers are also kept incase the
2322  *  super_block handler wishes to interpret them.
2323  */
2324 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2325 {
2326
2327         if (info->raid_disks == 0) {
2328                 /* just setting version number for superblock loading */
2329                 if (info->major_version < 0 ||
2330                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2331                     super_types[info->major_version].name == NULL) {
2332                         /* maybe try to auto-load a module? */
2333                         printk(KERN_INFO
2334                                 "md: superblock version %d not known\n",
2335                                 info->major_version);
2336                         return -EINVAL;
2337                 }
2338                 mddev->major_version = info->major_version;
2339                 mddev->minor_version = info->minor_version;
2340                 mddev->patch_version = info->patch_version;
2341                 return 0;
2342         }
2343         mddev->major_version = MD_MAJOR_VERSION;
2344         mddev->minor_version = MD_MINOR_VERSION;
2345         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2346         mddev->ctime         = get_seconds();
2347
2348         mddev->level         = info->level;
2349         mddev->size          = info->size;
2350         mddev->raid_disks    = info->raid_disks;
2351         /* don't set md_minor, it is determined by which /dev/md* was
2352          * openned
2353          */
2354         if (info->state & (1<<MD_SB_CLEAN))
2355                 mddev->recovery_cp = MaxSector;
2356         else
2357                 mddev->recovery_cp = 0;
2358         mddev->persistent    = ! info->not_persistent;
2359
2360         mddev->layout        = info->layout;
2361         mddev->chunk_size    = info->chunk_size;
2362
2363         mddev->max_disks     = MD_SB_DISKS;
2364
2365         mddev->sb_dirty      = 1;
2366
2367         /*
2368          * Generate a 128 bit UUID
2369          */
2370         get_random_bytes(mddev->uuid, 16);
2371
2372         return 0;
2373 }
2374
2375 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2376 {
2377         mdk_rdev_t *rdev;
2378
2379         rdev = find_rdev(mddev, dev);
2380         if (!rdev)
2381                 return -ENODEV;
2382
2383         md_error(mddev, rdev);
2384         return 0;
2385 }
2386
2387 static int md_ioctl(struct inode *inode, struct file *file,
2388                         unsigned int cmd, unsigned long arg)
2389 {
2390         char b[BDEVNAME_SIZE];
2391         int err = 0;
2392         struct hd_geometry *loc = (struct hd_geometry *) arg;
2393         mddev_t *mddev = NULL;
2394
2395         if (!capable(CAP_SYS_ADMIN))
2396                 return -EACCES;
2397
2398         /*
2399          * Commands dealing with the RAID driver but not any
2400          * particular array:
2401          */
2402         switch (cmd)
2403         {
2404                 case RAID_VERSION:
2405                         err = get_version((void *)arg);
2406                         goto done;
2407
2408                 case PRINT_RAID_DEBUG:
2409                         err = 0;
2410                         md_print_devices();
2411                         goto done;
2412
2413 #ifndef MODULE
2414                 case RAID_AUTORUN:
2415                         err = 0;
2416                         autostart_arrays(arg);
2417                         goto done;
2418 #endif
2419                 default:;
2420         }
2421
2422         /*
2423          * Commands creating/starting a new array:
2424          */
2425
2426         mddev = inode->i_bdev->bd_disk->private_data;
2427
2428         if (!mddev) {
2429                 BUG();
2430                 goto abort;
2431         }
2432
2433
2434         if (cmd == START_ARRAY) {
2435                 /* START_ARRAY doesn't need to lock the array as autostart_array
2436                  * does the locking, and it could even be a different array
2437                  */
2438                 static int cnt = 3;
2439                 if (cnt > 0 ) {
2440                         printk(KERN_WARNING
2441                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2442                                "This will not be supported beyond 2.6\n",
2443                                current->comm, current->pid);
2444                         cnt--;
2445                 }
2446                 err = autostart_array(new_decode_dev(arg));
2447                 if (err) {
2448                         printk(KERN_WARNING "md: autostart %s failed!\n",
2449                                 __bdevname(arg, b));
2450                         goto abort;
2451                 }
2452                 goto done;
2453         }
2454
2455         err = mddev_lock(mddev);
2456         if (err) {
2457                 printk(KERN_INFO
2458                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2459                         err, cmd);
2460                 goto abort;
2461         }
2462
2463         switch (cmd)
2464         {
2465                 case SET_ARRAY_INFO:
2466
2467                         if (!list_empty(&mddev->disks)) {
2468                                 printk(KERN_WARNING
2469                                         "md: array %s already has disks!\n",
2470                                         mdname(mddev));
2471                                 err = -EBUSY;
2472                                 goto abort_unlock;
2473                         }
2474                         if (mddev->raid_disks) {
2475                                 printk(KERN_WARNING
2476                                         "md: array %s already initialised!\n",
2477                                         mdname(mddev));
2478                                 err = -EBUSY;
2479                                 goto abort_unlock;
2480                         }
2481                         {
2482                                 mdu_array_info_t info;
2483                                 if (!arg)
2484                                         memset(&info, 0, sizeof(info));
2485                                 else if (copy_from_user(&info, (void*)arg, sizeof(info))) {
2486                                         err = -EFAULT;
2487                                         goto abort_unlock;
2488                                 }
2489                                 err = set_array_info(mddev, &info);
2490                                 if (err) {
2491                                         printk(KERN_WARNING "md: couldn't set"
2492                                                 " array info. %d\n", err);
2493                                         goto abort_unlock;
2494                                 }
2495                         }
2496                         goto done_unlock;
2497
2498                 default:;
2499         }
2500
2501         /*
2502          * Commands querying/configuring an existing array:
2503          */
2504         /* if we are initialised yet, only ADD_NEW_DISK or STOP_ARRAY is allowed */
2505         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY && cmd != RUN_ARRAY) {
2506                 err = -ENODEV;
2507                 goto abort_unlock;
2508         }
2509
2510         /*
2511          * Commands even a read-only array can execute:
2512          */
2513         switch (cmd)
2514         {
2515                 case GET_ARRAY_INFO:
2516                         err = get_array_info(mddev, (void *)arg);
2517                         goto done_unlock;
2518
2519                 case GET_DISK_INFO:
2520                         err = get_disk_info(mddev, (void *)arg);
2521                         goto done_unlock;
2522
2523                 case RESTART_ARRAY_RW:
2524                         err = restart_array(mddev);
2525                         goto done_unlock;
2526
2527                 case STOP_ARRAY:
2528                         err = do_md_stop (mddev, 0);
2529                         goto done_unlock;
2530
2531                 case STOP_ARRAY_RO:
2532                         err = do_md_stop (mddev, 1);
2533                         goto done_unlock;
2534
2535         /*
2536          * We have a problem here : there is no easy way to give a CHS
2537          * virtual geometry. We currently pretend that we have a 2 heads
2538          * 4 sectors (with a BIG number of cylinders...). This drives
2539          * dosfs just mad... ;-)
2540          */
2541                 case HDIO_GETGEO:
2542                         if (!loc) {
2543                                 err = -EINVAL;
2544                                 goto abort_unlock;
2545                         }
2546                         err = put_user (2, (char *) &loc->heads);
2547                         if (err)
2548                                 goto abort_unlock;
2549                         err = put_user (4, (char *) &loc->sectors);
2550                         if (err)
2551                                 goto abort_unlock;
2552                         err = put_user(get_capacity(mddev->gendisk)/8,
2553                                                 (short *) &loc->cylinders);
2554                         if (err)
2555                                 goto abort_unlock;
2556                         err = put_user (get_start_sect(inode->i_bdev),
2557                                                 (long *) &loc->start);
2558                         goto done_unlock;
2559         }
2560
2561         /*
2562          * The remaining ioctls are changing the state of the
2563          * superblock, so we do not allow read-only arrays
2564          * here:
2565          */
2566         if (mddev->ro) {
2567                 err = -EROFS;
2568                 goto abort_unlock;
2569         }
2570
2571         switch (cmd)
2572         {
2573                 case ADD_NEW_DISK:
2574                 {
2575                         mdu_disk_info_t info;
2576                         if (copy_from_user(&info, (void*)arg, sizeof(info)))
2577                                 err = -EFAULT;
2578                         else
2579                                 err = add_new_disk(mddev, &info);
2580                         goto done_unlock;
2581                 }
2582                 case HOT_GENERATE_ERROR:
2583                         err = hot_generate_error(mddev, new_decode_dev(arg));
2584                         goto done_unlock;
2585                 case HOT_REMOVE_DISK:
2586                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2587                         goto done_unlock;
2588
2589                 case HOT_ADD_DISK:
2590                         err = hot_add_disk(mddev, new_decode_dev(arg));
2591                         goto done_unlock;
2592
2593                 case SET_DISK_FAULTY:
2594                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2595                         goto done_unlock;
2596
2597                 case RUN_ARRAY:
2598                         err = do_md_run (mddev);
2599                         goto done_unlock;
2600
2601                 default:
2602                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2603                                 printk(KERN_WARNING "md: %s(pid %d) used"
2604                                         " obsolete MD ioctl, upgrade your"
2605                                         " software to use new ictls.\n",
2606                                         current->comm, current->pid);
2607                         err = -EINVAL;
2608                         goto abort_unlock;
2609         }
2610
2611 done_unlock:
2612 abort_unlock:
2613         mddev_unlock(mddev);
2614
2615         return err;
2616 done:
2617         if (err)
2618                 MD_BUG();
2619 abort:
2620         return err;
2621 }
2622
2623 static int md_open(struct inode *inode, struct file *file)
2624 {
2625         /*
2626          * Succeed if we can lock the mddev, which confirms that
2627          * it isn't being stopped right now.
2628          */
2629         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2630         int err;
2631
2632         if ((err = mddev_lock(mddev)))
2633                 goto out;
2634
2635         err = 0;
2636         mddev_get(mddev);
2637         mddev_unlock(mddev);
2638
2639         check_disk_change(inode->i_bdev);
2640  out:
2641         return err;
2642 }
2643
2644 static int md_release(struct inode *inode, struct file * file)
2645 {
2646         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2647
2648         if (!mddev)
2649                 BUG();
2650         mddev_put(mddev);
2651
2652         return 0;
2653 }
2654
2655 static int md_media_changed(struct gendisk *disk)
2656 {
2657         mddev_t *mddev = disk->private_data;
2658
2659         return mddev->changed;
2660 }
2661
2662 static int md_revalidate(struct gendisk *disk)
2663 {
2664         mddev_t *mddev = disk->private_data;
2665
2666         mddev->changed = 0;
2667         return 0;
2668 }
2669 static struct block_device_operations md_fops =
2670 {
2671         .owner          = THIS_MODULE,
2672         .open           = md_open,
2673         .release        = md_release,
2674         .ioctl          = md_ioctl,
2675         .media_changed  = md_media_changed,
2676         .revalidate_disk= md_revalidate,
2677 };
2678
2679 int md_thread(void * arg)
2680 {
2681         mdk_thread_t *thread = arg;
2682
2683         lock_kernel();
2684
2685         /*
2686          * Detach thread
2687          */
2688
2689         daemonize(thread->name, mdname(thread->mddev));
2690
2691         current->exit_signal = SIGCHLD;
2692         allow_signal(SIGKILL);
2693         thread->tsk = current;
2694
2695         /*
2696          * md_thread is a 'system-thread', it's priority should be very
2697          * high. We avoid resource deadlocks individually in each
2698          * raid personality. (RAID5 does preallocation) We also use RR and
2699          * the very same RT priority as kswapd, thus we will never get
2700          * into a priority inversion deadlock.
2701          *
2702          * we definitely have to have equal or higher priority than
2703          * bdflush, otherwise bdflush will deadlock if there are too
2704          * many dirty RAID5 blocks.
2705          */
2706         unlock_kernel();
2707
2708         complete(thread->event);
2709         while (thread->run) {
2710                 void (*run)(mddev_t *);
2711
2712                 wait_event_interruptible(thread->wqueue,
2713                                          test_bit(THREAD_WAKEUP, &thread->flags));
2714                 if (current->flags & PF_FREEZE)
2715                         refrigerator(PF_FREEZE);
2716
2717                 clear_bit(THREAD_WAKEUP, &thread->flags);
2718
2719                 run = thread->run;
2720                 if (run)
2721                         run(thread->mddev);
2722
2723                 if (signal_pending(current))
2724                         flush_signals(current);
2725         }
2726         complete(thread->event);
2727         return 0;
2728 }
2729
2730 void md_wakeup_thread(mdk_thread_t *thread)
2731 {
2732         if (thread) {
2733                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2734                 set_bit(THREAD_WAKEUP, &thread->flags);
2735                 wake_up(&thread->wqueue);
2736         }
2737 }
2738
2739 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
2740                                  const char *name)
2741 {
2742         mdk_thread_t *thread;
2743         int ret;
2744         struct completion event;
2745
2746         thread = (mdk_thread_t *) kmalloc
2747                                 (sizeof(mdk_thread_t), GFP_KERNEL);
2748         if (!thread)
2749                 return NULL;
2750
2751         memset(thread, 0, sizeof(mdk_thread_t));
2752         init_waitqueue_head(&thread->wqueue);
2753
2754         init_completion(&event);
2755         thread->event = &event;
2756         thread->run = run;
2757         thread->mddev = mddev;
2758         thread->name = name;
2759         ret = kernel_thread(md_thread, thread, 0);
2760         if (ret < 0) {
2761                 kfree(thread);
2762                 return NULL;
2763         }
2764         wait_for_completion(&event);
2765         return thread;
2766 }
2767
2768 void md_interrupt_thread(mdk_thread_t *thread)
2769 {
2770         if (!thread->tsk) {
2771                 MD_BUG();
2772                 return;
2773         }
2774         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
2775         send_sig(SIGKILL, thread->tsk, 1);
2776 }
2777
2778 void md_unregister_thread(mdk_thread_t *thread)
2779 {
2780         struct completion event;
2781
2782         init_completion(&event);
2783
2784         thread->event = &event;
2785         thread->run = NULL;
2786         thread->name = NULL;
2787         md_interrupt_thread(thread);
2788         wait_for_completion(&event);
2789         kfree(thread);
2790 }
2791
2792 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
2793 {
2794         if (!mddev) {
2795                 MD_BUG();
2796                 return;
2797         }
2798
2799         if (!rdev || rdev->faulty)
2800                 return;
2801
2802         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
2803                 mdname(mddev),
2804                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
2805                 __builtin_return_address(0),__builtin_return_address(1),
2806                 __builtin_return_address(2),__builtin_return_address(3));
2807
2808         if (!mddev->pers->error_handler)
2809                 return;
2810         mddev->pers->error_handler(mddev,rdev);
2811         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2812         md_wakeup_thread(mddev->thread);
2813 }
2814
2815 /* seq_file implementation /proc/mdstat */
2816
2817 static void status_unused(struct seq_file *seq)
2818 {
2819         int i = 0;
2820         mdk_rdev_t *rdev;
2821         struct list_head *tmp;
2822
2823         seq_printf(seq, "unused devices: ");
2824
2825         ITERATE_RDEV_PENDING(rdev,tmp) {
2826                 char b[BDEVNAME_SIZE];
2827                 i++;
2828                 seq_printf(seq, "%s ",
2829                               bdevname(rdev->bdev,b));
2830         }
2831         if (!i)
2832                 seq_printf(seq, "<none>");
2833
2834         seq_printf(seq, "\n");
2835 }
2836
2837
2838 static void status_resync(struct seq_file *seq, mddev_t * mddev)
2839 {
2840         unsigned long max_blocks, resync, res, dt, db, rt;
2841
2842         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
2843         max_blocks = mddev->size;
2844
2845         /*
2846          * Should not happen.
2847          */
2848         if (!max_blocks) {
2849                 MD_BUG();
2850                 return;
2851         }
2852         res = (resync/1024)*1000/(max_blocks/1024 + 1);
2853         {
2854                 int i, x = res/50, y = 20-x;
2855                 seq_printf(seq, "[");
2856                 for (i = 0; i < x; i++)
2857                         seq_printf(seq, "=");
2858                 seq_printf(seq, ">");
2859                 for (i = 0; i < y; i++)
2860                         seq_printf(seq, ".");
2861                 seq_printf(seq, "] ");
2862         }
2863         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
2864                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
2865                        "resync" : "recovery"),
2866                       res/10, res % 10, resync, max_blocks);
2867
2868         /*
2869          * We do not want to overflow, so the order of operands and
2870          * the * 100 / 100 trick are important. We do a +1 to be
2871          * safe against division by zero. We only estimate anyway.
2872          *
2873          * dt: time from mark until now
2874          * db: blocks written from mark until now
2875          * rt: remaining time
2876          */
2877         dt = ((jiffies - mddev->resync_mark) / HZ);
2878         if (!dt) dt++;
2879         db = resync - (mddev->resync_mark_cnt/2);
2880         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
2881
2882         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
2883
2884         seq_printf(seq, " speed=%ldK/sec", db/dt);
2885 }
2886
2887 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
2888 {
2889         struct list_head *tmp;
2890         loff_t l = *pos;
2891         mddev_t *mddev;
2892
2893         if (l >= 0x10000)
2894                 return NULL;
2895         if (!l--)
2896                 /* header */
2897                 return (void*)1;
2898
2899         spin_lock(&all_mddevs_lock);
2900         list_for_each(tmp,&all_mddevs)
2901                 if (!l--) {
2902                         mddev = list_entry(tmp, mddev_t, all_mddevs);
2903                         mddev_get(mddev);
2904                         spin_unlock(&all_mddevs_lock);
2905                         return mddev;
2906                 }
2907         spin_unlock(&all_mddevs_lock);
2908         if (!l--)
2909                 return (void*)2;/* tail */
2910         return NULL;
2911 }
2912
2913 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2914 {
2915         struct list_head *tmp;
2916         mddev_t *next_mddev, *mddev = v;
2917
2918         ++*pos;
2919         if (v == (void*)2)
2920                 return NULL;
2921
2922         spin_lock(&all_mddevs_lock);
2923         if (v == (void*)1)
2924                 tmp = all_mddevs.next;
2925         else
2926                 tmp = mddev->all_mddevs.next;
2927         if (tmp != &all_mddevs)
2928                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
2929         else {
2930                 next_mddev = (void*)2;
2931                 *pos = 0x10000;
2932         }
2933         spin_unlock(&all_mddevs_lock);
2934
2935         if (v != (void*)1)
2936                 mddev_put(mddev);
2937         return next_mddev;
2938
2939 }
2940
2941 static void md_seq_stop(struct seq_file *seq, void *v)
2942 {
2943         mddev_t *mddev = v;
2944
2945         if (mddev && v != (void*)1 && v != (void*)2)
2946                 mddev_put(mddev);
2947 }
2948
2949 static int md_seq_show(struct seq_file *seq, void *v)
2950 {
2951         mddev_t *mddev = v;
2952         sector_t size;
2953         struct list_head *tmp2;
2954         mdk_rdev_t *rdev;
2955         int i;
2956
2957         if (v == (void*)1) {
2958                 seq_printf(seq, "Personalities : ");
2959                 spin_lock(&pers_lock);
2960                 for (i = 0; i < MAX_PERSONALITY; i++)
2961                         if (pers[i])
2962                                 seq_printf(seq, "[%s] ", pers[i]->name);
2963
2964                 spin_unlock(&pers_lock);
2965                 seq_printf(seq, "\n");
2966                 return 0;
2967         }
2968         if (v == (void*)2) {
2969                 status_unused(seq);
2970                 return 0;
2971         }
2972
2973         if (mddev_lock(mddev)!=0)
2974                 return -EINTR;
2975         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
2976                 seq_printf(seq, "%s : %sactive", mdname(mddev),
2977                                                 mddev->pers ? "" : "in");
2978                 if (mddev->pers) {
2979                         if (mddev->ro)
2980                                 seq_printf(seq, " (read-only)");
2981                         seq_printf(seq, " %s", mddev->pers->name);
2982                 }
2983
2984                 size = 0;
2985                 ITERATE_RDEV(mddev,rdev,tmp2) {
2986                         char b[BDEVNAME_SIZE];
2987                         seq_printf(seq, " %s[%d]",
2988                                 bdevname(rdev->bdev,b), rdev->desc_nr);
2989                         if (rdev->faulty) {
2990                                 seq_printf(seq, "(F)");
2991                                 continue;
2992                         }
2993                         size += rdev->size;
2994                 }
2995
2996                 if (!list_empty(&mddev->disks)) {
2997                         if (mddev->pers)
2998                                 seq_printf(seq, "\n      %llu blocks",
2999                                         (unsigned long long)mddev->array_size);
3000                         else
3001                                 seq_printf(seq, "\n      %llu blocks",
3002                                         (unsigned long long)size);
3003                 }
3004
3005                 if (mddev->pers) {
3006                         mddev->pers->status (seq, mddev);
3007                         seq_printf(seq, "\n      ");
3008                         if (mddev->curr_resync > 2)
3009                                 status_resync (seq, mddev);
3010                         else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3011                                 seq_printf(seq, "       resync=DELAYED");
3012                 }
3013
3014                 seq_printf(seq, "\n");
3015         }
3016         mddev_unlock(mddev);
3017
3018         return 0;
3019 }
3020
3021 static struct seq_operations md_seq_ops = {
3022         .start  = md_seq_start,
3023         .next   = md_seq_next,
3024         .stop   = md_seq_stop,
3025         .show   = md_seq_show,
3026 };
3027
3028 static int md_seq_open(struct inode *inode, struct file *file)
3029 {
3030         int error;
3031
3032         error = seq_open(file, &md_seq_ops);
3033         return error;
3034 }
3035
3036 static struct file_operations md_seq_fops = {
3037         .open           = md_seq_open,
3038         .read           = seq_read,
3039         .llseek         = seq_lseek,
3040         .release        = seq_release,
3041 };
3042
3043 int register_md_personality(int pnum, mdk_personality_t *p)
3044 {
3045         if (pnum >= MAX_PERSONALITY) {
3046                 printk(KERN_ERR
3047                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3048                        p->name, pnum, MAX_PERSONALITY-1);
3049                 return -EINVAL;
3050         }
3051
3052         spin_lock(&pers_lock);
3053         if (pers[pnum]) {
3054                 spin_unlock(&pers_lock);
3055                 MD_BUG();
3056                 return -EBUSY;
3057         }
3058
3059         pers[pnum] = p;
3060         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3061         spin_unlock(&pers_lock);
3062         return 0;
3063 }
3064
3065 int unregister_md_personality(int pnum)
3066 {
3067         if (pnum >= MAX_PERSONALITY) {
3068                 MD_BUG();
3069                 return -EINVAL;
3070         }
3071
3072         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3073         spin_lock(&pers_lock);
3074         pers[pnum] = NULL;
3075         spin_unlock(&pers_lock);
3076         return 0;
3077 }
3078
3079 void md_sync_acct(mdk_rdev_t *rdev, unsigned long nr_sectors)
3080 {
3081         rdev->bdev->bd_contains->bd_disk->sync_io += nr_sectors;
3082 }
3083
3084 static int is_mddev_idle(mddev_t *mddev)
3085 {
3086         mdk_rdev_t * rdev;
3087         struct list_head *tmp;
3088         int idle;
3089         unsigned long curr_events;
3090
3091         idle = 1;
3092         ITERATE_RDEV(mddev,rdev,tmp) {
3093                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3094                 curr_events = disk_stat_read(disk, read_sectors) +
3095                                 disk_stat_read(disk, write_sectors) -
3096                                 disk->sync_io;
3097                 if ((curr_events - rdev->last_events) > 32) {
3098                         rdev->last_events = curr_events;
3099                         idle = 0;
3100                 }
3101         }
3102         return idle;
3103 }
3104
3105 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3106 {
3107         /* another "blocks" (512byte) blocks have been synced */
3108         atomic_sub(blocks, &mddev->recovery_active);
3109         wake_up(&mddev->recovery_wait);
3110         if (!ok) {
3111                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3112                 md_wakeup_thread(mddev->thread);
3113                 // stop recovery, signal do_sync ....
3114         }
3115 }
3116
3117
3118 void md_write_start(mddev_t *mddev)
3119 {
3120         if (!atomic_read(&mddev->writes_pending)) {
3121                 mddev_lock_uninterruptible(mddev);
3122                 if (mddev->in_sync) {
3123                         mddev->in_sync = 0;
3124                         del_timer(&mddev->safemode_timer);
3125                         md_update_sb(mddev);
3126                 }
3127                 atomic_inc(&mddev->writes_pending);
3128                 mddev_unlock(mddev);
3129         } else
3130                 atomic_inc(&mddev->writes_pending);
3131 }
3132
3133 void md_write_end(mddev_t *mddev)
3134 {
3135         if (atomic_dec_and_test(&mddev->writes_pending)) {
3136                 if (mddev->safemode == 2)
3137                         md_wakeup_thread(mddev->thread);
3138                 else
3139                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3140         }
3141 }
3142
3143 static inline void md_enter_safemode(mddev_t *mddev)
3144 {
3145         if (!mddev->safemode) return;
3146         if (mddev->safemode == 2 &&
3147             (atomic_read(&mddev->writes_pending) || mddev->in_sync ||
3148                     mddev->recovery_cp != MaxSector))
3149                 return; /* avoid the lock */
3150         mddev_lock_uninterruptible(mddev);
3151         if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3152             !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3153                 mddev->in_sync = 1;
3154                 md_update_sb(mddev);
3155         }
3156         mddev_unlock(mddev);
3157
3158         if (mddev->safemode == 1)
3159                 mddev->safemode = 0;
3160 }
3161
3162 void md_handle_safemode(mddev_t *mddev)
3163 {
3164         if (signal_pending(current)) {
3165                 printk(KERN_INFO "md: %s in immediate safe mode\n",
3166                         mdname(mddev));
3167                 mddev->safemode = 2;
3168                 flush_signals(current);
3169         }
3170         md_enter_safemode(mddev);
3171 }
3172
3173
3174 DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3175
3176 #define SYNC_MARKS      10
3177 #define SYNC_MARK_STEP  (3*HZ)
3178 static void md_do_sync(mddev_t *mddev)
3179 {
3180         mddev_t *mddev2;
3181         unsigned int currspeed = 0,
3182                  window;
3183         sector_t max_sectors,j;
3184         unsigned long mark[SYNC_MARKS];
3185         sector_t mark_cnt[SYNC_MARKS];
3186         int last_mark,m;
3187         struct list_head *tmp;
3188         sector_t last_check;
3189
3190         /* just incase thread restarts... */
3191         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3192                 return;
3193
3194         /* we overload curr_resync somewhat here.
3195          * 0 == not engaged in resync at all
3196          * 2 == checking that there is no conflict with another sync
3197          * 1 == like 2, but have yielded to allow conflicting resync to
3198          *              commense
3199          * other == active in resync - this many blocks
3200          */
3201         do {
3202                 mddev->curr_resync = 2;
3203
3204                 ITERATE_MDDEV(mddev2,tmp) {
3205                         if (mddev2 == mddev)
3206                                 continue;
3207                         if (mddev2->curr_resync &&
3208                             match_mddev_units(mddev,mddev2)) {
3209                                 printk(KERN_INFO "md: delaying resync of %s"
3210                                         " until %s has finished resync (they"
3211                                         " share one or more physical units)\n",
3212                                        mdname(mddev), mdname(mddev2));
3213                                 if (mddev < mddev2) {/* arbitrarily yield */
3214                                         mddev->curr_resync = 1;
3215                                         wake_up(&resync_wait);
3216                                 }
3217                                 if (wait_event_interruptible(resync_wait,
3218                                                              mddev2->curr_resync < mddev->curr_resync)) {
3219                                         flush_signals(current);
3220                                         mddev_put(mddev2);
3221                                         goto skip;
3222                                 }
3223                         }
3224                         if (mddev->curr_resync == 1) {
3225                                 mddev_put(mddev2);
3226                                 break;
3227                         }
3228                 }
3229         } while (mddev->curr_resync < 2);
3230
3231         max_sectors = mddev->size << 1;
3232
3233         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3234         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3235                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3236         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3237                "(but not more than %d KB/sec) for reconstruction.\n",
3238                sysctl_speed_limit_max);
3239
3240         is_mddev_idle(mddev); /* this also initializes IO event counters */
3241         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3242                 j = mddev->recovery_cp;
3243         else
3244                 j = 0;
3245         for (m = 0; m < SYNC_MARKS; m++) {
3246                 mark[m] = jiffies;
3247                 mark_cnt[m] = j;
3248         }
3249         last_mark = 0;
3250         mddev->resync_mark = mark[last_mark];
3251         mddev->resync_mark_cnt = mark_cnt[last_mark];
3252
3253         /*
3254          * Tune reconstruction:
3255          */
3256         window = 32*(PAGE_SIZE/512);
3257         printk(KERN_INFO "md: using %dk window, over a total of %Lu blocks.\n",
3258                 window/2,(unsigned long long) max_sectors/2);
3259
3260         atomic_set(&mddev->recovery_active, 0);
3261         init_waitqueue_head(&mddev->recovery_wait);
3262         last_check = 0;
3263
3264         if (j)
3265                 printk(KERN_INFO
3266                         "md: resuming recovery of %s from checkpoint.\n",
3267                         mdname(mddev));
3268
3269         while (j < max_sectors) {
3270                 int sectors;
3271
3272                 sectors = mddev->pers->sync_request(mddev, j, currspeed < sysctl_speed_limit_min);
3273                 if (sectors < 0) {
3274                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3275                         goto out;
3276                 }
3277                 atomic_add(sectors, &mddev->recovery_active);
3278                 j += sectors;
3279                 if (j>1) mddev->curr_resync = j;
3280
3281                 if (last_check + window > j)
3282                         continue;
3283
3284                 last_check = j;
3285
3286                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3287                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3288                         break;
3289
3290         repeat:
3291                 if (jiffies >= mark[last_mark] + SYNC_MARK_STEP ) {
3292                         /* step marks */
3293                         int next = (last_mark+1) % SYNC_MARKS;
3294
3295                         mddev->resync_mark = mark[next];
3296                         mddev->resync_mark_cnt = mark_cnt[next];
3297                         mark[next] = jiffies;
3298                         mark_cnt[next] = j - atomic_read(&mddev->recovery_active);
3299                         last_mark = next;
3300                 }
3301
3302
3303                 if (signal_pending(current)) {
3304                         /*
3305                          * got a signal, exit.
3306                          */
3307                         printk(KERN_INFO
3308                                 "md: md_do_sync() got signal ... exiting\n");
3309                         flush_signals(current);
3310                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3311                         goto out;
3312                 }
3313
3314                 /*
3315                  * this loop exits only if either when we are slower than
3316                  * the 'hard' speed limit, or the system was IO-idle for
3317                  * a jiffy.
3318                  * the system might be non-idle CPU-wise, but we only care
3319                  * about not overloading the IO subsystem. (things like an
3320                  * e2fsck being done on the RAID array should execute fast)
3321                  */
3322                 mddev->queue->unplug_fn(mddev->queue);
3323                 cond_resched();
3324
3325                 currspeed = ((unsigned long)(j-mddev->resync_mark_cnt))/2/((jiffies-mddev->resync_mark)/HZ +1) +1;
3326
3327                 if (currspeed > sysctl_speed_limit_min) {
3328                         if ((currspeed > sysctl_speed_limit_max) ||
3329                                         !is_mddev_idle(mddev)) {
3330                                 current->state = TASK_INTERRUPTIBLE;
3331                                 schedule_timeout(HZ/4);
3332                                 goto repeat;
3333                         }
3334                 }
3335         }
3336         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3337         /*
3338          * this also signals 'finished resyncing' to md_stop
3339          */
3340  out:
3341         mddev->queue->unplug_fn(mddev->queue);
3342
3343         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3344
3345         /* tell personality that we are finished */
3346         mddev->pers->sync_request(mddev, max_sectors, 1);
3347
3348         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3349             mddev->curr_resync > 2 &&
3350             mddev->curr_resync > mddev->recovery_cp) {
3351                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3352                         printk(KERN_INFO
3353                                 "md: checkpointing recovery of %s.\n",
3354                                 mdname(mddev));
3355                         mddev->recovery_cp = mddev->curr_resync;
3356                 } else
3357                         mddev->recovery_cp = MaxSector;
3358         }
3359
3360         md_enter_safemode(mddev);
3361  skip:
3362         mddev->curr_resync = 0;
3363         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3364         md_wakeup_thread(mddev->thread);
3365 }
3366
3367
3368 /*
3369  * This routine is regularly called by all per-raid-array threads to
3370  * deal with generic issues like resync and super-block update.
3371  * Raid personalities that don't have a thread (linear/raid0) do not
3372  * need this as they never do any recovery or update the superblock.
3373  *
3374  * It does not do any resync itself, but rather "forks" off other threads
3375  * to do that as needed.
3376  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3377  * "->recovery" and create a thread at ->sync_thread.
3378  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3379  * and wakeups up this thread which will reap the thread and finish up.
3380  * This thread also removes any faulty devices (with nr_pending == 0).
3381  *
3382  * The overall approach is:
3383  *  1/ if the superblock needs updating, update it.
3384  *  2/ If a recovery thread is running, don't do anything else.
3385  *  3/ If recovery has finished, clean up, possibly marking spares active.
3386  *  4/ If there are any faulty devices, remove them.
3387  *  5/ If array is degraded, try to add spares devices
3388  *  6/ If array has spares or is not in-sync, start a resync thread.
3389  */
3390 void md_check_recovery(mddev_t *mddev)
3391 {
3392         mdk_rdev_t *rdev;
3393         struct list_head *rtmp;
3394
3395
3396         dprintk(KERN_INFO "md: recovery thread got woken up ...\n");
3397
3398         if (mddev->ro)
3399                 return;
3400         if ( ! (
3401                 mddev->sb_dirty ||
3402                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3403                 test_bit(MD_RECOVERY_DONE, &mddev->recovery)
3404                 ))
3405                 return;
3406         if (mddev_trylock(mddev)==0) {
3407                 int spares =0;
3408                 if (mddev->sb_dirty)
3409                         md_update_sb(mddev);
3410                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3411                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3412                         /* resync/recovery still happening */
3413                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3414                         goto unlock;
3415                 }
3416                 if (mddev->sync_thread) {
3417                         /* resync has finished, collect result */
3418                         md_unregister_thread(mddev->sync_thread);
3419                         mddev->sync_thread = NULL;
3420                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3421                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3422                                 /* success...*/
3423                                 /* activate any spares */
3424                                 mddev->pers->spare_active(mddev);
3425                         }
3426                         md_update_sb(mddev);
3427                         mddev->recovery = 0;
3428                         /* flag recovery needed just to double check */
3429                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3430                         wake_up(&resync_wait);
3431                         goto unlock;
3432                 }
3433                 if (mddev->recovery) {
3434                         /* probably just the RECOVERY_NEEDED flag */
3435                         mddev->recovery = 0;
3436                         wake_up(&resync_wait);
3437                 }
3438
3439                 /* no recovery is running.
3440                  * remove any failed drives, then
3441                  * add spares if possible
3442                  */
3443                 ITERATE_RDEV(mddev,rdev,rtmp) {
3444                         if (rdev->raid_disk >= 0 &&
3445                             rdev->faulty &&
3446                             atomic_read(&rdev->nr_pending)==0) {
3447                                 mddev->pers->hot_remove_disk(mddev, rdev->raid_disk);
3448                                 rdev->raid_disk = -1;
3449                         }
3450                         if (!rdev->faulty && rdev->raid_disk >= 0 && !rdev->in_sync)
3451                                 spares++;
3452                 }
3453                 if (mddev->degraded) {
3454                         ITERATE_RDEV(mddev,rdev,rtmp)
3455                                 if (rdev->raid_disk < 0
3456                                     && !rdev->faulty) {
3457                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3458                                                 spares++;
3459                                         else
3460                                                 break;
3461                                 }
3462                 }
3463
3464                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3465                         /* nothing we can do ... */
3466                         goto unlock;
3467                 }
3468                 if (mddev->pers->sync_request) {
3469                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3470                         if (!spares)
3471                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3472                         mddev->sync_thread = md_register_thread(md_do_sync,
3473                                                                 mddev,
3474                                                                 "%s_resync");
3475                         if (!mddev->sync_thread) {
3476                                 printk(KERN_ERR "%s: could not start resync"
3477                                         " thread...\n",
3478                                         mdname(mddev));
3479                                 /* leave the spares where they are, it shouldn't hurt */
3480                                 mddev->recovery = 0;
3481                         } else {
3482                                 md_wakeup_thread(mddev->sync_thread);
3483                         }
3484                 }
3485         unlock:
3486                 mddev_unlock(mddev);
3487         }
3488 }
3489
3490 int md_notify_reboot(struct notifier_block *this,
3491                                         unsigned long code, void *x)
3492 {
3493         struct list_head *tmp;
3494         mddev_t *mddev;
3495
3496         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3497
3498                 printk(KERN_INFO "md: stopping all md devices.\n");
3499
3500                 ITERATE_MDDEV(mddev,tmp)
3501                         if (mddev_trylock(mddev)==0)
3502                                 do_md_stop (mddev, 1);
3503                 /*
3504                  * certain more exotic SCSI devices are known to be
3505                  * volatile wrt too early system reboots. While the
3506                  * right place to handle this issue is the given
3507                  * driver, we do want to have a safe RAID driver ...
3508                  */
3509                 mdelay(1000*1);
3510         }
3511         return NOTIFY_DONE;
3512 }
3513
3514 struct notifier_block md_notifier = {
3515         .notifier_call  = md_notify_reboot,
3516         .next           = NULL,
3517         .priority       = INT_MAX, /* before any real devices */
3518 };
3519
3520 static void md_geninit(void)
3521 {
3522         struct proc_dir_entry *p;
3523
3524         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3525
3526         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3527         if (p)
3528                 p->proc_fops = &md_seq_fops;
3529 }
3530
3531 int __init md_init(void)
3532 {
3533         int minor;
3534
3535         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3536                         " MD_SB_DISKS=%d\n",
3537                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3538                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3539
3540         if (register_blkdev(MAJOR_NR, "md"))
3541                 return -1;
3542         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3543                 unregister_blkdev(MAJOR_NR, "md");
3544                 return -1;
3545         }
3546         devfs_mk_dir("md");
3547         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3548                                 md_probe, NULL, NULL);
3549         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3550                             md_probe, NULL, NULL);
3551
3552         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3553                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3554                                 S_IFBLK|S_IRUSR|S_IWUSR,
3555                                 "md/%d", minor);
3556
3557         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3558                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3559                               S_IFBLK|S_IRUSR|S_IWUSR,
3560                               "md/d%d", minor);
3561
3562
3563         register_reboot_notifier(&md_notifier);
3564         raid_table_header = register_sysctl_table(raid_root_table, 1);
3565
3566         md_geninit();
3567         return (0);
3568 }
3569
3570
3571 #ifndef MODULE
3572
3573 /*
3574  * Searches all registered partitions for autorun RAID arrays
3575  * at boot time.
3576  */
3577 static dev_t detected_devices[128];
3578 static int dev_cnt;
3579
3580 void md_autodetect_dev(dev_t dev)
3581 {
3582         if (dev_cnt >= 0 && dev_cnt < 127)
3583                 detected_devices[dev_cnt++] = dev;
3584 }
3585
3586
3587 static void autostart_arrays(int part)
3588 {
3589         char b[BDEVNAME_SIZE];
3590         mdk_rdev_t *rdev;
3591         int i;
3592
3593         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3594
3595         for (i = 0; i < dev_cnt; i++) {
3596                 dev_t dev = detected_devices[i];
3597
3598                 rdev = md_import_device(dev,0, 0);
3599                 if (IS_ERR(rdev)) {
3600                         printk(KERN_ALERT "md: could not import %s!\n",
3601                                 __bdevname(dev, b));
3602                         continue;
3603                 }
3604                 if (rdev->faulty) {
3605                         MD_BUG();
3606                         continue;
3607                 }
3608                 list_add(&rdev->same_set, &pending_raid_disks);
3609         }
3610         dev_cnt = 0;
3611
3612         autorun_devices(part);
3613 }
3614
3615 #endif
3616
3617 static __exit void md_exit(void)
3618 {
3619         mddev_t *mddev;
3620         struct list_head *tmp;
3621         int i;
3622         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3623         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3624         for (i=0; i < MAX_MD_DEVS; i++)
3625                 devfs_remove("md/%d", i);
3626         for (i=0; i < MAX_MD_DEVS; i++)
3627                 devfs_remove("md/d%d", i);
3628
3629         devfs_remove("md");
3630
3631         unregister_blkdev(MAJOR_NR,"md");
3632         unregister_blkdev(mdp_major, "mdp");
3633         unregister_reboot_notifier(&md_notifier);
3634         unregister_sysctl_table(raid_table_header);
3635         remove_proc_entry("mdstat", NULL);
3636         ITERATE_MDDEV(mddev,tmp) {
3637                 struct gendisk *disk = mddev->gendisk;
3638                 if (!disk)
3639                         continue;
3640                 export_array(mddev);
3641                 del_gendisk(disk);
3642                 put_disk(disk);
3643                 mddev->gendisk = NULL;
3644                 mddev_put(mddev);
3645         }
3646 }
3647
3648 module_init(md_init)
3649 module_exit(md_exit)
3650
3651 EXPORT_SYMBOL(register_md_personality);
3652 EXPORT_SYMBOL(unregister_md_personality);
3653 EXPORT_SYMBOL(md_error);
3654 EXPORT_SYMBOL(md_sync_acct);
3655 EXPORT_SYMBOL(md_done_sync);
3656 EXPORT_SYMBOL(md_write_start);
3657 EXPORT_SYMBOL(md_write_end);
3658 EXPORT_SYMBOL(md_handle_safemode);
3659 EXPORT_SYMBOL(md_register_thread);
3660 EXPORT_SYMBOL(md_unregister_thread);
3661 EXPORT_SYMBOL(md_wakeup_thread);
3662 EXPORT_SYMBOL(md_print_devices);
3663 EXPORT_SYMBOL(md_interrupt_thread);
3664 EXPORT_SYMBOL(md_check_recovery);
3665 MODULE_LICENSE("GPL");